CH 12

12
Information
Asymmetry and
Agency Theory
. . . the single largest cost that
stands between issuers and
investors is the problem of
information asymmetry.
Bernard S. Black, Journal of Small and
Emerging Business Law 2, 1998, 9199.
The divorce of ownership from
control . . . almost necessarily
involves a new form of economic
organization of society.
Adolph A. Berle and Gardiner C. Means,
The Modern Corporation and Private
Property, Transaction Publishers, 1991.
I
NFORMATION ASYMMETRY AND AGENCY THEORY play a central role in corporate
nance and have their roots in the information economics literature. Information asymmetry
occurs when one group of participants has better or more timely information than other groups.
A signal is an action taken by the more informed that provides credible information to the less
informed. Typically, the source of the information asymmetry is the superior knowledge that
managers have about the rms prospects, while the investors in the rms comprise the uninformed
group. Agency theory derives from the fact that decisions within rms are made by management,
who are agents for the investors. Conicting interests between management and investors can lead
to suboptimal allocation of resources within the rm. As stated by Lambert [2001], agency theory
evaluates the impact of the conict of interest between principals and agents because of (1) shirking
by the agent, (2) diversion of resources by the agent for private consumption, (3) differential time
horizon of the agent and principal, and (4) differential risk aversion of the agent and the principal.
The purpose of this chapter is to provide a foundation for the role of information asymmetry and
agency theory in corporate nance. Specically, signaling provides testable empirical implications
for rm decisions about capital structure, dividend policy, new investments, and stock splits.
Agency theory provides its own explanation for the same decisions. Empirical testing is still trying
to wring out the separable implications of these two fundamental theories.
A. Information Asymmetry
This section provides a discussion of information asymmetry and signaling. Specically, consider a
world in which there are two types of rmslowand high quality. The managers of the high-quality
rms would like to signal their superiority to the market. These signals can either have exogenous
costs associated with them (costly signals) or have costs that are endogenous (costless or cheap-
talk signals). Examples of such signals include the level of investment in the rm, the amount of
415
416 Chapter 12: Information Asymmetry and Agency Theory
debt issued, the size of the dividend declared, the type of nancing used for an investment, and
the decision to split the stock. Regardless of the choice of the signaling mechanism, high-quality
rms can separate themselves from low-quality rms as long as the low-quality rm cannot mimic
the actions of a high-quality rm. Such a separating equilibrium requires that the gain to the low-
quality rms from mimicking must be lower than the cost associated with signaling falsely. If the
cost is lower than the gain, low-quality rms would mimic the signaling actions of high-quality
rms, thus leading to a pooling equilibrium where the market would be unable to distinguish
between them.
1. Costly Signaling
One of the rst papers that explicitly related uncertainty with price and quality is Akerlof [1970].
He examines the market for automobiles where there are four types of cars availablenewor used
and good or bad. In his framework, individuals buy new cars without knowing whether they are
good or bad. On the other hand, after owning the car for some time, the owners get a better idea of
the quality of the car; that is, information asymmetry develops since the owners (potential sellers)
have more knowledge about the car than the potential buyers. Since the potential buyers cannot
tell the difference between good and bad cars, they are willing to offer the same amount for both
and, therefore, the good and bad cars sell for the same price. As a result, a potential seller of a
used car cannot receive the true value since the true quality is not known to the potential buyer. In
addition, since the seller would always nd it benecial to sell a bad car and buy a new one if the
price of a used car is more than the expected value of a new car, this would imply that a used car
will never be priced above this expected value. As a result, it is predicted that good cars may not
be traded at all while bad cars will drive the good cars out of the market.
Spence [1973] extends the logic of Akerlofs argument by formally examining a market in
which signaling takes place, there are a relatively large number of signalers, and the signalers do
not acquire signaling reputation. He demonstrates the existence of a signaling equilibrium with
a specic example in the context of job market signaling. Specically, there are two groups of
job seekers within the population and they both face one employer. The two groups differ in
productivity, with that of group A being 1 and that of group B being 2. Group A represents a
fraction q of the population and a potential signal (for example, years of education) is available
at a cost. The cost for group A to acquire y units of education is y (that is, $1 per unit), while the
cost for group B is 0.5y.
Nowsuppose that the employer believes that there is a level of education, y, such that if y <y,
she expects the employees productivity to be equal to 1 with probability one, and if y y, then
the productivity will be equal to 2 with probability one. Therefore, his wage schedule W(y) would
be 1 if y < y and would be 2 if y y.
In this framework, each group will select their education level at either 0 or y, since acquiring
education between 0 and y or beyond y does not provide any marginal benets (wage) but has
a marginal cost. For this to be a separating equilibrium with group A selecting an education level
of 0 and group B selecting a level of y, we must have 1> 2 y and 2
y
2
> 1 or 1< y < 2.
This follows from the fact that if group A (B) selects an education of 0, their gain is 1 (1), while
the gain is 2 y(2
y
2
) if an education of y is selected. It should be noted that there are an
innite number of equilibrium values for y since for any y in the range of 1 to 2, the employer
can perfectly distinguish between the two groups.
Spence then proceeds to show that group A is worse off with the existence of signaling since
they would be paid more than 1
_
q +2(1q) =2 q
_
in the no-signaling case. On the other
A. Information Asymmetry 417
hand, group B is not always better off with the signaling. For example, if 50% of the population
is group 1 (q is 0.5), then the no-signaling wage for both groups is 1.5 (with no associated cost)
while the signaling wage for group B is 2 at a cost of
y
2
. Therefore the net benet to group B
from signaling is 0.5
y
2
. Since y > 1, this implies the net benet from signaling is negative,
and group B would prefer a solution without signaling. Finally Spence shows that under certain
circumstances (for example, different wage functions), pooling equilibria exist with both groups
either selecting an education level of 0 or an education level of y. In such pooling equilibria,
education levels convey no useful information.
Using a similar framework, Rothschild and Stiglitz [1976] and Wilson [1978] analyze a model
of information transformation in insurance markets. They show that in a market with two classes
of customers, high risk and low risk, there is a separating equilibrium where the two groups
buy different insurance contracts. Specically they show that rather than offering a single price
per unit of coverage, rms have incentives to charge a higher price for increased coverage. This
causes high-risk individuals to be separated fromlow-risk individuals since the additional coverage
yields greater marginal benets for the high-risk individuals but not for low-risk individuals.
They also show that a Nash equilibrium may also fail to exist.
1
Riley [1975, 1979] conrms
that a Nash equilibrium does not exist for this class of models with a continuum of types of
informed participants. Riley also provides formal conditions under which signaling equilibria exist.
Specically he shows that the multiple signaling equilibria in the Spence model reduce to a single
equilibrium that is Pareto optimal.
Applications of the costly signaling model of Spence to issues in corporate nance include
Leland and Pyle [1977], Ross [1977], and Bhattacharya [1979, 1980] in the context of entrepre-
neurs seeking nancing for projects that only they know the value of, managers of rms signaling
their rms better prospects with debt nancing, and managers signaling rm cash ows by pre-
committing to higher levels of dividends, respectively.
SIGNALING PROJECT QUALITY WITH INVESTMENT In the Leland and Pyle [1977]
analysis, an entrepreneur plans to invest in a project by retaining fraction of the projects equity
and raising the remainder from other investors. The project involves a capital outlay of $X and a
future value of +, where is the expected end-of-period value and is a random variable with
a mean of 0 and a standard deviation of
. The entrepreneur has information that allows her to

assign a particular value to the expected end-of-period value but has no credible way to convey this
information to other investors who have a probability distribution for . However, these investors
are assumed to respond to a signal fromthe entrepreneur regarding her knowledge of . The signal
is the fraction of the project that the entrepreneur retains, ; that is, investors perceive that is a
function of .
In this framework, the market value of the project, V, given a signal , is
V() =
1
(1+r)
[() ], (12.1)
1
A Nash equilibrium is a set of strategies for participants in a noncooperative game. The equilibrium strategies have the
property that no participant can do better than choose her equilibriumstrategy holding the strategies of all other participants
xed. Therefore, no participant would want to change her strategy once she has seen what the other participants have done.
For example, consider a two-person game in which the persons (A and B) can take either action x or y. If they both take
action x (y), they each receive a payoff of $4 (2). If one takes action x and the other takes action y, the payoffs are $1 for
the one choosing x and $3 for the one choosing y. This game has two Nash equilibriaone in which they both choose
action x and the other in which they both choose action y.
where
r =the risk-free rate of return,
() =the valuation schedule used by the market to infer the expected
end-of-period value from the signal, ,
=the markets adjustment for the risk of the project.
The entrepreneur is assumed to maximize her expected utility of terminal wealth subject to a
budget constraint; that is,
maximize E
_
U
_
W
1
__
(12.2)
subject to W
0
=X +V
M
+Y (1)V(), (12.3)
where
W
0
=the entrepreneurs initial wealth,
V
M
=the value of the market portfolio,
=the fraction of the market portfolio owned by the entrepreneur,
Y =the amount invested in the risk-free asset,
=the fraction of the project the entrepreneur retains.
The uncertain end of period wealth of the entrepreneur,

W
1
, is given by
W
1
= ( +) +

M +(1+r)Y
= [ + () +] +
_
M (1+r)V
M
_
+(1+r)(W
0
X) +() (12.4)
and

M is the gross return of the market portfolio.
2
The rst-order conditions for the maximization imply that
E
_
U
_
W
1
__
=E
_
U
W
1
_
[ + () + +(1)
]
_
=0 (12.5)
and
E
_
U
_
W
1
__
=E
_
U
W
1
_
[
M (1+r)V
M
]
_
=0, (12.6)
where
.
2
The second equality in the denition of

W
1
is obtained by substituting the denitions of V() and Y (from the denition
of W
0
).
Assuming the existence of an equilibrium valuation schedule, () =, Eq. (12.5) can be
written as
(1)
=
E
_
U
W
1
_
[ +]
_
E
_
U
W
1
__ . (12.7)
Equation (12.6) can be used to solve for the entrepreneurs optimal holdings of the market portfolio,
(, ). Substituting this relation for in Eq. (12.7) provides a differential equation relating
to with any equilibrium schedule satisfying this differential equation.
Leland and Pyle proceed to show, based on the condition dened by Eq. (12.7) and the second-
order conditions for the maximization, that for entrepreneurs with normal demand for equity in the
project (1) the equilibriumvaluation schedule, () is strictly increasing in and (2) in equilibrium
with signaling through , an entrepreneur would hold a larger fraction of the project as compared
to the case that they can communicate the true value costlessly to investors.
3
The rst result implies
that an entrepreneur with a higher-value project will retain a larger share of the project compared
to an entrepreneur with a low-value project. The second result suggests that entrepreneurs suffer
a welfare loss by being forced to hold a fraction of equity that is larger than what they would
have optimally held if the value of the project could be communicated costlessly. The cost of the
signal represents this welfare loss. This is a separating equilibrium since the gain to a entrepreneur
of a low-value project of signaling falselythat is, retaining a larger fraction of the project than
implied by Eq. (12.7)is smaller than the welfare loss sustained by deviating even more from the
costless communication holding.
Grinblatt and Hwang [1989] generalize the Leland and Pyle model by assuming that both the
mean and the variance of the projects cash ows are unknown. As a result, the fraction of the project
that the entrepreneur retains is not sufcient by itself to signal the expected value of the project.
A second signal is needed to infer the variance of the projects cash ows since the equilibrium
signaling schedule is a function of both the variance and the fraction retained by the entrepreneur.
This second signal is observed when the offering price of the issue is announced and is the degree of
underpricing per share. The Grinblatt and Hwang model has a number of empirical implications,
with some of them being consistent with the Leland and Pyle model. Specically, their model
predicts that (1) holding the degree of underpricing constant, the variance of a projects cash
ows is negatively related to the fraction retained by the entrepreneur, (2) holding the fraction
retained by the entrepreneur constant, the value of the project is positively related to its variance,
(3) holding the projects variance constant, its value is positively related to the fraction retained by
the entrepreneur, (4) holding the projects value constant, the variance of a projects cash ows is
negatively related to the fraction retained by the entrepreneur, (5) holding the fraction retained by
the entrepreneur constant, the degree of underpricing is positively related to its variance, (6) holding
the projects variance constant, the degree of underpricing is positively related to the fraction
retained by the entrepreneur, (7) holding the fraction retained by the entrepreneur constant, the
value of the project is positively related to the degree of underpricing, and (8) holding the projects
variance constant, its value is positively related to the degree of underpricing.
SIGNALING FIRM QUALITY WITH DEBT Ross [1977] considers a market with two types
of rms, A and B, in a two-date world.
4
At time 1, type A rms have a value, V
a
(=100), greater
3
The second-order conditions are

2
E[]
2
< 0,

2
E[]
2
< 0, and
_
2
E[]
2
_ _
2
E[]
2
_
2
E[]
_
2
> 0.
4
A discussion of the Ross model is also contained in Chapter 15.
than type B rms with value V
b
(=50) < V
a
. If there is no uncertainty in the market and pricing
is risk neutral, the time 0 values of the two types of rms would be given by
V
0a
=
V
1a
1+r
=
100
1
=100 (12.8)
and
V
0b
=
V
1b
1+r
=
50
1
=50 < V
0a
, (12.9)
where
r = the risk-free interest rate (=0%).
Now suppose there is uncertainty and investors cannot differentiate between the two types of
rms. If q(=0.4) is the proportion of type A rms and investors assume that rms are type A with
probability q and type B with probability (1q), then all rms in the market will have the same
value given by
V
0
=
qV
1a
+(1q)V
1b
1+r
=
(0.4)(100) +(0.6)(50)
1
=70. (12.10)
In this framework, if type Arms would attempt to signal that they are of type A, Brms would give
the same signal, resulting in no discrimination between rms in the market (a pooling equilibrium).
Ross suggests that one way to resolve this issue (that is, create a separating equilibrium) is
to assume that the manager of a rm is accountable for nancing decisions made at time 0.
Specically, assume that managers know the true quality of their rms, they are not allowed to
trade in their rms securities, they issue debt at time 0, they are compensated by an incentive
schedule that is known to investors, and they act to maximize their incentive compensation. The
compensation schedule is given by
M =(1+r)
0
V
0
+
1
_
V
1
if V
1
> D
V
1
C if V
1
D
, (12.11)
where
0
,
1
are positive weights, V
1
is the value of the rm at time 1, D is the face value of debt
issued by the rm at time 0, and C is a penalty imposed on the manager if V
1
< D. For example,
assume that
0
=0.1 and
1
=0.2. Thus the managers compensation is
M =0.1V
0
+0.15
_
V
1
if V
1
> D
V
1
C if V
1
D
. (12.12)
Ross shows that the compensation schedule as dened in Eq. (12.11) can be used to establish
a Spence-style signaling equilibrium. Assume that D
is the maximum amount of debt a type B

rm can carry without going into bankruptcy. Further assume that if D > D
, investors perceive
the rm to be of type A and if D D
, investors perceive the rm to be of type B. For this to be

established as a signaling equilibrium, the signal has to be unambiguous and managers must have
the incentive to always issue the correct signal, that is, to tell the truth rather than lie.
Based on the signal (debt level) chosen by the manager of a type A rm, her compensation
would be
M
A
(D) =
_
0
V
1a
+
1
V
1a
if D
< D V
1a
0
V
1b
+
1
V
1a
if D D
=
_
25 if D
< D V
1a
20 if D D
. (12.13)
The manager of a type A rm would have the incentive to issue the correct signal (choose a debt
level higher then D
) as long as her compensation from signaling correctly is greater than her

compensation based on an false signal. In this case, since the marginal payoff from telling the
truth is greater than that from a lie, that is, (
0
V
1a
+
1
V
1a
) =25 >20 =(
0
V
1b
+
1
V
1a
), she will
give the correct signal.
The compensation of the manager of a type B rm is given by
M
B
(D) =
_
0
V
1a
+
1
(V
1b
C) if D
< D V
1a
0
V
1b
+
1
V
1b
if D D
=
_
17.5 0.15C if D
< D V
1a
12.5 if D D
. (12.14)
Again, the manager of a type B rm will have the incentive to signal correctly if
[
0
V
1a
+
1
(V
1b
C)] = 17.5 0.15C < 12.5 = (
0
V
1b
+
1
V
1b
) or if
0
(V
1a
V
1b
) = 5 <
0.15C =
1
C or if C > 33.33. Therefore, the managers of type B rms would signal correctly
if their marginal gain is less than the cost they bear for signaling falsely. In the example we have
used, the manager of the type B rms chooses to signal correctly if the cost imposed on him for
lying is larger than 33.33. As in the Spence paper, there are multiple equilibrium values for D
in this case. The main empirical implication of the Ross model is that rms with larger expected
future cash ows should issue more debt.
Guedes and Thompson [1995] develop a model based on the Ross model in which the choice
between xed-rate and oating-rate debt serves as a signal of rm quality. In their model, costs of
nancial distress provide an incentive for managers to choose borrowing strategies that stabilize net
income. They showthat a separating equilibriumexists where rms above a minimumquality issue
high default risk debt and those below this minimum issue low default risk debt. This equilibrium,
in conjunction with the result that there is a unique threshold for the volatility of expected ination
at which xed- and variable-rate debt have the same default risk, implies that xed-rate nancing
is a favorable signal above the volatility threshold, while variable-rate debt is a favorable signal
below the volatility threshold.
The empirical evidence on the use of debt as a signal of rm quality is mixed. Studies that have
focused on decisions by rms to change leverage and analyzed the impact of the announcement
of these decisions on stock value have found evidence consistent with the signaling role of debt.
Specically, they nd that leverage-increasing transactions are associated with increases in stock
price while the opposite is true for leverage-decreasing transactions. On the other hand, many
cross-sectional studies have found that rm protability is negatively related to debt, indicating
that more protable rms carry less debt. This result is not consistent with higher-quality rms
being associated with larger amounts of debt. A more detailed review of the empirical evidence
on capital structure is provided in Chapter 15.
SIGNALINGEXPECTEDCASHFLOWS WITHDIVIDENDS Bhattacharya [1979] develops
a dividend-signaling model in which the liquidation value of a rmis related to the actual dividend
paid.
5
In his model, a rm is considering a new (perpetual) project with end-of-period cash ows
5
A discussion of the Bhattacharya model is also provided in Chapter 16.
denoted by X. The rm signals information about the cash ows associated with the project
by declaring an incremental dividend of D. The signaling response of the incremental dividend
commitment is an incremental liquidation value of V(D). If the project cash ow is larger than the
committed dividend (X > D), then the rm can reduce the amount of external nancing it needs
by the amount (X D). If, on the other hand, the cash ow is less than the dividend (X <D), the
dividend is still paid but the rm faces a shortfall. Bhattacharya assumes that the cost of making
up the shortfall is more than the benet of the cash ow surplus because of costs associated with
raising external funds. Specically, he assumes that in the case of a shortfall the cost to the rm
(shareholders) is (1+) [X D]. Finally, he also assumes that shareholders pay a personal tax
of
p
on dividends and pay no tax on capital gains.
In this framework, the shareholders receive the incremental liquidation value that results from
the response to the signal, V(D), and the after-tax value of the committed dividend, (1
P
)D.
In addition, if the cash ow, X, is more than the committed dividend, the shareholders receive
the residual, X D. On the other hand, if the cash ow is less than the committed dividend, the
shareholders face a shortfall of (1+)(X D). Therefore, the incremental objective function of
the shareholders is
E(D) =
1
1+r
_
V(D) +(1
P
)D +
_
X
D
(X D)f (X)dX +
_
D
X
(1+)(X D)f (X)dX
_
=
1
(1+r)
_
V(D) +
P
D
_
D
X
(X D)f (X)dX
_
, (12.15)
where r is the after-tax rate of interest, the cash ow X is distributed over the range
_
X, X
_
,
and is the expected cash ow. The intuitive explanation of Eq. (12.15) is that the value to the
shareholder is the sum of the incremental liquidation value and the expected cash ow less the
amount of taxes paid on the dividend received and the cost borne to fund the cash ow shortfall.
Managers choose D to maximize E(D) given a market signaling value function V(D). Now
consider a project with cash ows uniformly distributed over the range [0, t ]. Substituting the
density function for this uniform distribution in Eq. (12.15) implies that managers choose D to
maximize
E(D) =
1
1+r
_
V(D) +
t
2

P
D
D
2
2t
_
. (12.16)
The rst-order condition associated with this maximization is
V
(D
) =
P
+
D
t
. (12.17)
Equation (12.17) states that at the optimal dividend, the marginal benet from signaling has to
equal its marginal cost. The marginal benet, V
(D
), is the change in the liquidation value from

a unit change in the dividend, while the marginal cost,
P
+
D
t
, is the sum of the marginal tax
rate (the tax paid for a unit change in dividend) and the change in the cost borne to fund a cash
ow shortfall.
The market signaling value function V(D) is an equilibrium schedule only if V[D
(t )] is
the true value of future cash ows for the project whose cash ows are being signaled by the
dividend commitment of D
(t ). Given the assumptions that the project is perpetual, the dividend

is stationary, and that no learning takes place over time, then the equilibrium market signaling
value function is given by
V
_
D
(t )
_
=
1
r
_
t
2

P
D
(t )
_
D
(t )
_
2
2t
_
. (12.18)
Equations (12.17) and (12.18) can be solved for the equilibrium D
(t ) and V(D) schedules.

Specically, they imply that
D
(t ) =At (12.19)
and
V
_
D
(t )
_
=
_
P
+A
_
D
(t ), (12.20)
where A determines the response of value, V(D), to the committed dividend, D, and is given by
A =
_
_ _
1+r
1+2r
_
+
_
_ _
1+r
1+2r
_
_
1+
(1+2r)
2
P
(1+r)
2
. (12.21)
Bhattacharyas results imply that the equilibrium response of value, V(D), to the committed
dividend, D, namely, the value of A, is a decreasing function of the personal dividend tax rate,
P
,
the cost associated with a cash ow shortfall, , and the rate of interest, r. The rst two results
follow from the argument that if for a higher tax rate or cash ow shortfall cost, V(D) responded
to
P
or only, the optimizing dividend will be the same. As a result, V(D) would overestimate
the true value of future cash ows. This implies that, in equilibrium, A would have to be lower.
Similarly, if the rate of interest increases, the present value of future cash ows decreases. Thus,
in equilibrium this requires a lower response of V(D) to D, implying a lower A.
To illustrate these results numerically, consider a situation in which the personal tax rate is
40% (=
P
), the shortfall penalty is 60% (=), the upper bound of the cash ow is $100 (=t )
and the after-tax rate of interest is 25%. In this case the value of A is 0.64, the optimal dividend
is to pay $64, and the value response to this dividend is $50. If the tax rate is increased to 50%,
the corresponding gures are 0.57, $57, and $48, respectively. On the other hand, decreasing the
shortfall penalty to 40% increases A to 0.7, increases the dividend to $70, but decreases the value
response to $48. Finally a decrease in interest rates to 10% increases all three variables, with A
being .72, the dividend being $72, and the value response being $59.
John and Williams [1985] provide an alternate model for the use of dividends as a signal of
private information held by insiders. They consider a two-date (one-period) model in which insiders
of an all-equity rm commit to an investment of I at time 0. Conditional on this investment, they
select a dividend D. Funds for the investment and dividends are raised by cash held by the rm
(C) or selling new shares of stock (N) at the ex-dividend price per share (p
e
). Thus the sources
and uses of funds satisfy
I +D =C +Np
e
=C +P
e
, (12.22)
where P
e
=Np
e
. Dividends are costly to stockholders in the sense that they have to pay tax on
the dividend at the marginal personal tax rate of
P
. At time 1, each rm realizes its cash ows,
and stockholders receive a liquidating dividend. Shareholders do not pay taxes on the liquidating
dividend. The present value of the future cash ows is denoted by X, 1X <. Insiders have
private information about X that they are attempting to signal with the selection of the aggregate
dividend D. If there are Q shares of stock outstanding prior to the issue of new equity, the cum-
dividend price per share (p) and the ex-dividend price per share (p
e
) have to be related by
p =p
e
+
(1
P
)D
Q
to preclude arbitrage. Finally, it is assumed that shareholders have a demand for liquidity (L), and
they meet this demand through the dividend and the sale of M shares of stock at the ex-dividend
price of p
e
, that is, L = D + p
e
M. In this framework, insiders select the optimal dividend to
maximize the rms true value to its current stockholders, that is,
maximize
D
_
(1
P
)D +p
e
M +
QM
Q+N
X
_
. (12.23)
In the above expression, the rst term is the after-tax dividend receipt, the second term represents
the process of the sale of M shares of stock at the ex-dividend price of p
e
, and the last term is
the value of the QM shares they are left with after the sale. Substitution of the source and use
of funds, and the no-arbitrage and the liquidity constraints, into Eq. (12.23) yields the alternate
objective function
maximize
D
L
P
D +
_
P +
P
D L
P +
P
D +I C
_
X, (12.24)
where P =Qp. The rst-order condition for this maximization is
P
=
_
P
+
P
D
_
L +I C
_
P +
P
D +I C
_ X. (12.25)
The rst-order condition in Eq. (12.25) states that at the optimal dividend the marginal cost
of the dividend to the shareholder (
P
) is equal to the marginal benet to the current shareholder
from signaling. Solving this rst-order condition along with the normalizing assumption that the
dividend for the rm with the most unfavorable information (X =1) is zero yields the following
optimal dividend:
D(X) =
1
P
max (I C +L, 0) ln X. (12.26)
As can be seen from Eq. (12.26), the optimal dividend increases in the present value of the cash
ows and shareholders liquidity demands and decreases in the personal tax rate and the supply
of cash. John and Williams also show that the market value of the rms stock is the net value of
the rm minus the optimal signaling costs, that is,
P[D(X)] =C +X I
P
D(X), (12.27)
and the impact of announced increments in dividends is
P
D
=
P
P[D(X)] +
P
D(X) L
I C +L
. (12.28)
Thus increments in dividends cause an increase in market price.
Empirical evidence on the relation between dividend changes and stock values and future
earnings changes provides support for the signaling role of dividends. Specically, it has been
reported that the announcement of increases in dividends is associated with stock price increases
while the reverse holds true for dividend decreases. In addition, it has been shown that the
announcements of dividend initiations are associated with increases in stock price. Finally, when
tracking earnings announcements two years following a dividend increase, studies have reported
unexpected positive earnings changes. Chapter 16 provides a more detailed discussion on the
empirical evidence associated with dividend policy.
SIGNALING AND THE ISSUE-INVEST DECISION Myers and Majluf [1984] consider a
three-date model (time 1, 0, and 1) of a rm that has assets in place and a valuable investment
opportunity (project).
6
At time 1, the market has the same information about the assets in place
and the project as the management of the rm. Specically, both management and the market know
the distributions of the future value of the assets in place (
A) and the net present value (NPV) of the

project (
B). At time 0, management receives additional information about the value of the assets
in place and the NPV of the project. In particular, they observe the realization of

A (=a) and

B
(=b). The market receives this information at time +1. The project requires an investment of I and
can be nanced by issuing stock, selling marketable securities (short-term assets of the company),
and/or drawing down on the rms holding of cash. The total amount of cash that can be obtained
from the last two options is S and is referred to as nancial slack. The amount of slack available
is known at time 0 to the market. It is further assumed that the investment required is greater than
the nancial slack available (S < I). Thus investing in the project requires an equity issue of E
(=I S) at time 0. Management is assumed to act in the best interest of shareholders who own
stock at time 0 by maximizing the value of the old shares conditional on the issue-invest decision
and their knowledge of a and b, that is, maximize V
old
0
=V(a, b, E). Since the market does not
know the values of a and b at time 0, the market value of their shares will not necessarily be equal
to V
old
0
.
If the management of the rm decides not to issue new equity, it foregoes the project and the
value of the old shares would be
V
old
0
=S +a. (12.29)
If, on the other hand, management issues equity and invests in the project, the value of the old
shares would be
V
old
0
=
P
+E
(E +S +a +b), (12.30)
where P
is the market price of old shares if stock is issued.

The old shareholders will not be worse off under the condition that
P
+E
(E +S +a +b) S +a (12.31)
6
A discussion of the Myers-Majluf model is also provided in Chapter 15.
or when
P
+E
(E +b)
E
P
+E
(S +a)
(E +b)
E
P
(S +a) . (12.32)
The condition specied in Eq. (12.32) states that management should issue and invest for
combinations of a and b such that the gain to the old shareholders from investing is not less
than the fraction of the no-invest value that is captured by new shareholders.
7
Specically, the
lower the value of a and the higher the value of b, the more the rm is likely to invest. This
condition also implies that under certain circumstances, a rm may give up on good investment
opportunities rather than issuing new equity to raise funds. The loss in share value as a result of
this is L =P(do not issue and invest)E(
B|do not issue and invest). This loss decreases with the
slack (S) and increases with the investment required (I) and the amount of equity that needs to be
raised to fund the investment (E).
If a is known to all investors and managers, then stock is always issued as long as b 0. To see
this, consider the share price if a is known. Specically, it is given by
P
=S +a +E(
B|issue and invest) (12.33)

Since E(
B|issue and invest) > 0, this implies that P
> S +a or
S+a
P
< 1 or
E
P
(S +a) < E. If
b 0, then E +b > E. Combining these two inequalities yields
E
P
(S +a) < E < E +b. Equa-
tion (12.32) always holds if a is known to all market participants and b 0. Therefore, the rm
will always issue equity and invest if faced with a nonnegative NPV project.
If the rmhas no investment opportunities available, this model suggests that the rmwill issue
and invest only in bad states of the world. To see this, consider a situation where a has a lower
bound of a
min
and all market participants know that a cannot be lower than a
min
. This implies that
the price of old shares cannot be less than S +a
min
. Assume that P
=S +a
min
+. Substituting
this expression for P
into Eq. (12.32) yields the condition that the rm issues equity only if
S +a
S +a
min
+
< 1
or if a < a
min
+. This implies that E(
A|issue and invest) < a

min
+. Combining this condi-
tion with the denition of P
= S + a
min
+ yields P
> S + E(
A|issue and invest), a con-

tradiction to the denition of the price of the stock. Therefore if the rm has no investment
opportunities available, P
=S +a
min
. Substituting this expression for P
in the condition dened

by Eq. (12.32) implies that the rm issues equity only if
S +a
S +a
min
1. (12.34)
Equation (12.34) only holds if a =a
min
. Thus with no investment opportunities, the rm issues
equity only if the value of assets in place is at its lower bound.
7
The price P
is given by P
=S +E(
A|issue and invest) +E(
B|issue and invest).

In the Myers-Majluf model, the decision to issue new equity always reduces stock prices
unless the rm is going to issue new equity with a probability of one. To see this, consider the
condition in Eq. (12.32). The rmchooses not to issue equity if there exist combinations of a and b
such that
(E +b) <
E
P
(S +a) or (12.35)
a > P
_
1+
b
E
_
S.
Since
b
E
0, the rm would choose not to issue equity as long as a > P
S or P
< S +a. If
the rm decides not to issue new equity, the share price is given by P =S +E(
A|do not issue

and invest) S +a >P
. Therefore, the price must fall when the rm chooses to issue and invest.
Intuitively, this result follows from the fact that the decision to issue signals to the market that the
realization of

A is such that the value of assets in place are in a region with relatively low values
for a.
To consider the potential impact of debt nancing on this equilibrium, assume that the rm
can raise the funding it needs with debt or equity, and that the nancing policies are announced at
time 1 and adhered to at time 0. Suppose the rm issues equity. Then the value with the equity
issue is V
old
issue
=a +b +I E
1
, where E
1
is the value of the newly issued shares at time 1. Since
I =S +E, this implies that
V
old
issue
=S +a +b
_
E
1
E
_
=S +a +b E =V
old
no issue
+b E.
Since the rm will choose to issue only if V
old
issue
V
old
no issue
, this implies that for the issue equity
and invest decision to be made, the projects NPV should not be lower than the gain to the new
shareholders, that is, b E. On the other hand, if the rm issues debt to nance the project, the
same logic leads to the conclusion that the rm would issue debt and invest as long as b D. If
debt is risk-free, then D =0 and the rm issues risk-free debt and invests as long as the project
has nonnegative NPV, that is, b 0. If debt is risky, then D is not zero but |D| <|E|. This
implies that if a rm chooses to issue equity it will also be willing (and, in fact, would prefer) to
issue debt. In addition, when D < b < E, the rm would issue debt but not issue equity. This
suggests that rms would follow a pecking order in nancinginternal funds followed by risk-
free debt, risky debt, and equity. The pecking order theory of capital structure and empirical
tests of it are also discussed in Chapter 15.
Krasker [1986] extends the Myers and Majluf model to allow rms to choose the size of the
new investment project and the equity issue. In addition to obtaining the same result as Myers and
Majluf, he also shows that the larger the equity issue the worse the signal.
Cooney and Kalay [1993] extend the Myers and Majluf model to allow for the existence of
negative NPV projects. Using the same structure as the Myers and Majluf model, they argue that
the value of the old shares if the rm issues and invests can be either greater or smaller than the
preannouncement value. As a result the decision to issue does not always result in a stock price
decrease. Specically, if the preannouncement value of the old stock is the weighted average of the
value if the rm issued and invested and the value if the rm did not issue, where the weights
are the probabilities of issuing and not issuing, respectively, then the announcement to issue will
be associated with a price increase if and only if the value if the rm issued equity is larger than
the value if the rm did not issue. If the rm is restricted to nonnegative NPV projects, this will
never hold true. On the other hand, if the rm is faced with both nonnegative and negative NPV
projects, Cooney and Kalay show that there are combinations of a and b for which the value if
the rm issued is larger than the value if the rm did not issue, and the announcement of the issue
would be associated with an increase in stock price.
Stein [1992] uses an adaptation of the Myers and Majluf model to examine the role of convertible
bonds in the pecking order theory of capital structure. He considers a three-date model (time 0, 1,
and 2) with three types of rms (good, medium, and bad). Each rm has access to a project with
required investment at time 0 of I and expected net present value of B. The discount rate is assumed
to be zero, all agents are risk neutral, the amount required for investment has to be raised from
external sources, and the rm is completely owned by its manager prior to the infusion of capital.
Each rm receives a cash ow from the investment of either b
L
or b
H
at time 2 (b
L
< I < b
H
).
Firms differ in the ex ante probability of receiving b
H
, with good types receiving b
H
with certainty,
medium types receiving b
H
with probability p, and bad types receiving b
H
with probability q
(q <p <1). Firm types are private information at time 0, and the true value of bad rms is volatile
between time 0 and time 1. At time 1, the rm type is revealed, and for bad rms, the value of the
probability q is updated to either 0 or p. The probability of deterioration is assumed to be z and
for consistency q =(1 z)p. The rm has three nancing options at time 0straight debt that
matures at time 2, convertible debt that matures at time 2 but can be called to force conversion at
time 1 at a predetermined conversion ratio, and equity. Debt nancing is associated with a potential
for costly nancial distress where a deadweight cost of c is imposed on the owner-manager.
Stein shows that if the costs of nancial distress are such that c > (I b
L
), then there is a
separating equilibriumin which good rms issued debt with face value I, bad rms issue a fraction
I
qb
H
+(1q)b
L
of equity, and medium rms issue convertible bonds with face value F > b
L
, a call price K,
b
L
< K < I, and convertible to
I
pb
H
+(1p)b
L
of equity. For this to be a separating equilibrium, the rms should not want to mimic each other. For
example, consider a situation in which the bad rm issues convertible debt. If it does so, there is
a probability of z that the rm will deteriorate and will, therefore, not be able to force conversion
at time 1. This follows from the fact that if the rm deteriorates, the conversion value of the
bond is
Ib
L
pb
H
+(1p)b
L
,
which is below the call price K. Since the cash ow for the deteriorated bad rm is b
L
with
probability 1 and this cash owis less than the face value of debt (F), the bad rmwould be forced
into nancial distress at time 2 with probability z, and the expected cost of distress would be zc. On
the other hand, the bad rm would be issuing an overpriced security at time 0 and would receive
a gain from the overpricing. Specically, the bad rm raises I with a security that will become a
straight debt claim with probability z and become an equity claim with probability (1 z). The
payoff on the security would be b
L
and I, respectively. This implies that the expected payoff is
zb
L
+(1z)I and the overpricing is z(I b
L
) [=I (zb
L
+(1z)I)]. This overpricing is less
than the expected cost of nancial distress, that is, z(I b
L
) < zc, since it has been assumed that
c >(I b
L
). Therefore, a bad rmwill not mimic a mediumrm. Using a similar argument it can
be shown that a bad rm will not mimic a good rm by issuing straight debt, a medium rm will
not want to mimic a bad or good rm, and there is no reason for a good rm to deviate from the
policy of issuing straight debt. Thus, a convertible debt issue allows a medium rm to get equity
into its capital structure while conveying positive news to the market.
The pecking order hypothesis predicts that, holding investment constant, leverage should de-
crease with protability since more protable rms have more access to internal capital. Similarly,
leverage should increase with investments, holding protability constant. Consistent with the peck-
ing order hypothesis, a number of empirical studies have reported a negative relation between
leverage and protability. In addition, Fama and French [2002] report that short-term variations in
investments are absorbed by changes in debt, a result consistent with the pecking order hypothesis.
On the other hand, they also report that rms with higher investments have less leverage. Minton
and Wruck [2002] nd for a group of conservatively nanced rms that they do not exhaust all
internal funds before they seek external funds. Lemmon and Zender [2002] nd for a sample of
rms that are likely to gain most from debt nancing that they are no less likely to issue equity
when seeking outside nancing. These results are not consistent with the pecking order hypothesis.
COSTLY SIGNALING AND STOCK SPLITS Copeland and Brennan [1988] consider a two-
period world where a stock split may reveal private information held by management about the
future prospects of the rm. Specically, assume that the manager of an all-equity rm has private
information at time 0 about its true expected future cash ows at time 1, X. At time 0, the rm
has m shares outstanding, and the manager can announce a split factor, s. Therefore, after the split
the number of shares outstanding becomes n =ms. At time 0, the value of the rm as assessed by
investors is denoted by P(z) if no split is announced and

P(n, z) if a split is announced, where
z is a vector of observable rm characteristics (e.g., cash ows) that affect value. At time 1, the
value of the cash ows, P, is revealed, and the value at which the rm will trade is P T (m, P)
if no split is announced and P T (n, P) if a split is announced. T (a, B), the transaction cost
incurred by the shareholder when there are a shares outstanding and the value of the cash ows
is B, is dened as t
1
B +t
2
a
B
1
with t
1
, t
2
> 0 and > 1.
The wage paid to the manager is assumed to be a linear function of the markets assessment of
value, the value itself, and the transactions costs faced by the shareholder.
8
Specically, if a split
is not announced, the managers wage is given by
W
0
(z) =P(z) +P T (m, P). (12.36)
On the other hand, the managers wage with a split announcement is
W
S
(n, z) =
P(n, z) +P T (n, P). (12.37)

The manager will decide to split only if W
S
(n
, z) > W
0
(z), where n
maximizes the wage as

dened in Eq. (12.37). The rst-order condition for the maximization is
W
n
=
P
n
t
2
P
1
=0. (12.38)
8
Copeland [1979] documents higher transactions costs per dollar of transaction for lower-priced stocks.
Since the managers private information is fully revealed by the split announcement, the value
assessed by investors after the split announcement must equal true value, that is,

P(n, z) =P.
Substituting this consistency condition in Eq. (12.38) yields the following differential equation for
the market assessment:
P
n
P
1
=t
2
. (12.39)
The solution to Eq. (12.39) is
P(n, z) =k[n +c(z)]

1/
, (12.40)
where k =
_
t
2
/
_
1/
and c(z) is a constant of integration. Therefore, the market value of a rm
after it announces a split is
M(n, z) =

P(n, z) T (n, P) =k(1t
1
)[n +c(z)]
1/
t
2
nk
1
[n +c(z)]
(1 )/
. (12.41)
Equation (12.41) implies that the value of the rm increases in the number of shares outstanding.
This follows from the fact that even though both the intrinsic value and the total transactions costs
increase with the number of shares outstanding, the incremental transactions cost due to an increase
in the number of shares outstanding is lower than the incremental intrinsic value.
Equation (12.41) has a number of testable empirical implications. Specically, consider the
situation where c(z) =0, that is, where the constant of integration is zero. In that case, Eq. (12.41)
can be written as
M(n, z) =
_
k(1t
1
) t
2
k
1
_
n
1/
. (12.42)
Equation (12.42) implies that the return after the announcement of a split can be written as
u =M(n, z)/M(z) =
_
k(1t
1
) t
2
k
1
_
n
1/
/M(z) =K
_
n
M(z)
_
1/ _
M(z)
_
1/ 1
, (12.43)
where M(z) is the presplit value of the rm and K =
_
k(1t
1
) t
2
k
1
_
.
Taking the log of Eq. (12.43) yields
ln u =ln K
_
1
_
ln
_
M(z)
n
_
+
_
1
1
_
ln
_
M(z)
_
. (12.44)
Equation (12.44) implies that a regression of the return associated with the announcement of a stock
split should be a decreasing function of the logs of the target share price
_
=
M(z)
n
_
and the presplit
market capitalization of the stock (=M(z)). In other words, the announcement date return for a
stock that splits 2 for 1 from $100 per share to $50 will be smaller than the announcement date
return of another rm that splits 2 for 1 from $30 per share to $15. The equation also implies
that the coefcients of the two variables should sum to 1. Copeland and Brennan estimate
Eq. (12.44) for 967 stock splits over the period 1967 to 1976. They nd that the coefcient of
the target share price is negative and signicant while that of market capitalization is negative
but not signicant. They also reject the hypothesis that the two coefcients sum to 1. They also
report that these two variables explain about 15% of the variation in announcement returns. Their
model also predicts that the growth in earnings postsplit for rms that actually decided to split
will be higher than for similar rms that decided not to split. McNichols and Dravid [1990] report
results that are consistent with this prediction. Overall, these results support the prediction of the
signaling model that the number of shares outstanding after the split provides new information to
investors.
2. Costless Signaling (Cheap Talk)
In the previous section, we considered signaling models that have exogenously specied signaling
costs. Crawford and Sobel [1982] consider a model in which they consider a costless message
(cheap talk) as a signal. Specically they consider a situation in which a sender has private
information in that she observes the value of a random variable (information type) and sends a
signal about this to a receiver. The receiver, in turn, takes an action in response to the signal.
This action then determines the payoff to both sender and receiver. The equilibrium in this model
consists of a family of signaling rules for the sender and action rules for the receiver such that
(1) the senders signaling rule yields an expected utility maximizing action for each of her
information types taking receivers action rules as given and (2) the receiver responds optimally to
each possible signal using Bayess lawto update his priors taking account of the senders signaling
strategy and the signal he receives. The model, therefore, has an endogenous signaling cost since
the utility of the sender can be affected by action taken by the receiver in response to the signal
he receives. Crawford and Sobel demonstrate the existence of equilibria where this endogenous
signaling cost is such that it is optimal for the sender to tell the truth.
In an extension of the Crawford and Sobel model, Austen-Smith and Banks [2000] also allow
the sender to have the ability to accept some direct loss in utility to transmit information in a more
credible manner (costly signals) than employing cheap talk (costless signals) alone. They showthat
the set of equilibria can be dramatically increased when costly signals can be used. They also show
that the availability of costly signals can improve the precision of cheap talk communications.
COSTLESS SIGNALING WITH STOCK SPLITS Brennan and Hughes [1991] consider a
situation where the manager of a rm has private information about its future cash ows and
wants to communicate it to the market. They assume that there is no credible and costless way to
communicate the information. They consider a situation in which managers use splits to change
stock price since these splits affect the incentives of brokerage houses to provide earnings forecasts
that reveal the managers private information.
9
Brennan and Hughes have a four-date world where at time 0 investors and managers have
homogenous prior beliefs about the rms future value, X. At time 1, the manager receives private
information about X and announces the number of new shares, n, through a stock split. At time 2,
N analysts gather information and announce earnings forecasts. At time 3, cash ow/value is
realized and analysts are paid their commission. All individuals are assumed to be risk neutral and
the priors on X are that it is normally distributed with mean X
0
and variance 1/s
0
(precision s
0
).
Assuming a zero interest rate, the market value of the rm at time 0 is
V
0
=(1t )X
0
, (12.45)
9
It is not clear, however, what mechanism explains just how brokerage houses discover private information held by
managers.
where t is the brokerage commission rate that is assumed to depend on stock price. At time 1,
the manager receives a noisy signal about the rms value, Y
m
=X +
m
, where
m
is normally
distributed with mean zero and precision s
m
. The manager then announces n, and investors infer
that the managers signal was

Y
m
(n) and revise their beliefs about X to
E(X|n) =
X(n) =
X
0
s
0
+
Y
m
(n)s
m
s
0
+s
m
. (12.46)
The new market value of the rm is
V
1
(n) =
X(n) T (n) C, (12.47)

where C is the cost of executing the split and T (n) is the expected total brokerage commission
(=E
_
Xt (X/n)|n
_
).
Brennan and Hughes argue that if T (n) is monotonic, the information contained in the an-
nouncement of n is equivalent to that contained in the announcement of T . Therefore, they assume
that the manager announces T , and the market value of the rmafter the managerial announcement
can be written as
V
1
(n) =
X(T ) T C, (12.48)
where

X(T ) =E(X|T ). If the cost to an individual analyst of making an earnings forecast is f ,
the number of analysts who make forecasts is
N(T ) =T/f =FT, (12.49)
where F =1/f . The forecast of an analyst i is Y
i
=X +
i
, where
i
is normally distributed with
mean 0 and precision s. The value of the rmafter the forecasts have been released will be given by
V
2
(T, Y) =
X
0
s
0
+
Y
m
s
m
+YFT s
s
0
+s
m
+FT s
E
_
Xt (X/n)|T,

Y
m
, Y
_
C, (12.50)
where Y is the average value of Y
i
. Equation (12.50) indicates that as the number of analysts
making forecasts increases, the greater the weight attached to the average analysts forecast. Thus
managers with good news would be motivated to attract the attention of more analysts.
The objective of the manager is to choose n or T at time 1 in order to maximize the expected
value of the rm at time 2, with this expectation being conditioned on the managers private
information. Specically, the manager wants to maximize
E[V
2
(T )|Y
m
] =
X
0
s
0
+
Y
m
s
m
+
_
X
0
s
0
+Y
m
s
m
s
0
+s
m
_
FT s
s
0
+s
m
+FT s
T C, (12.51)
since
E
_
Y|Y
m
_
=
X
0
s
0
+Y
m
s
m
s
0
+s
m
and
E
__
Xt (X/n)|T,

Y
m
, Y
_
|Y
m
_
=T.
The rst-order condition for this maximization along with the consistency condition that
Y
m
=
Y
m
(T ) yields the following differential equation for the investors valuation schedule:
m
(T ) =
s
0
+s
m
+FT s
s
m
. (12.52)
The solution to this differential equation is
Y
m
(T ) =
s
0
+s
m
s
m
T +
Fs
2s
m
T
2
+K, (12.53)
where K is a constant of integration. The investors valuation schedule is an increasing function
of the aggregate brokerage commission.
To summarize, in this section we presented a model developed by Brennan and Hughes in
which managers with private information have an incentive to attract the attention of analysts
so that they will discover the value of this private information and transmit it to investors using
earnings forecasts. Specically, the manager achieves this by announcing a stock split that reduces
stock price and thereby increasing brokerage commissions that result from the research conducted
by the analysts in the brokerage house. Knowing this, investors interpret stock splits as a signal
that the manager has favorable information.
Two direct empirical implications of this model are that the number of analysts following a
rm should be negatively related to share price and the change in the number of analysts should
be positively related to the magnitude of the stock split. Both these implications are empirically
supported by the data.
COSTLESS SIGNALING WITH DEBT AND EQUITY Heinkel [1982] develops a costless
separating equilibrium in which the amount of debt used by a rm is monotonically related to
its unobservable rm value.
10
In his model, each rm consists of a single one-period project that
requires nancing. The optimal investment amount is the same for all projects, and nancing is
available from perfectly competitive debt and equity markets. Denote V as the value of the project
that requires an investment of I, D as the face value of debt issued to fund the project, B as
the current value of debt, E as the value of equity issued to fund the project, S as the market
value of equity after the nancing and investment is completed, and as the proportion of equity
that is retained by insiders. Insiders are assumed to know the random function that will generate
the projects future cash ows, while outsiders (capital suppliers) know only the distribution of
functions across the economy, with each return generating function referenced by the quality rating
of the rm, n. It is further assumed that the value of the project is decreasing in n, (dV/dn < 0),
the value of debt is increasing in n (B/n >0), and
2
B/nD 0. These conditions imply that
high-n (high-quality) rms have safer debt but are less valuable. Therefore, high-quality (low-
quality) rms can benet from misrepresentation in the debt (equity) market.
10
See Chapter 15 for a more detailed description of the theory and empirical evidence on capital structure.
Insiders know the true value of the rms debt claim, B =B(n, D), and the value of the equity
position they retain, (D)[V(n) B(n, D)]. They choose D to maximize this value. The rst-
order condition for this optimization is
(d/dD) [V(n) B(n, D)] (D) (B/D) =0. (12.54)
The second-order condition is
_
d
2
/dD
2
_
[V(n) B(n, D)] 2(d/dD) (B/D) (D)
_
2
B/D
2
_
< 0. (12.55)
For this signaling equilibrium to be costless, insiders of a quality n rm should receive the net
present value of the project, V(n) I, or
_
D
(n)
_ _
V(n) B
_
n, D
(n)
__
[V(n) I] =0, (12.56)
where D
(n) is the optimal value of D for insiders of a rm of quality n as determined by

Eqs. (12.54) and (12.55).
Equations (12.54), (12.55), and (12.56) dene a costless signaling equilibrium in which the
insiders determine D
(n) and the market interprets the signal according to Eq. (12.48). It can be
shown that Eqs. (12.54) and (12.55) imply that low-quality rms issue more debt than high-quality
rms, that is, dD
(n)/dn < 0. To see this, take the total differential of Eq. (12.55) to get
_
d
2
dD
2
(V B) 2
d
dD
B
D

2
B
D
2
_
dD
dn
=
d
dD
_
B
n
dV
dn
_
+
_

2
B
nD
_
. (12.57)
Since it has been assumed that B/n > 0, V/n < 0,
2
B/nD 0, and d/dD > 0
from Eq. (12.54), the right-hand side of Eq. (12.57) is positive. In addition, since the brack-
eted term on the left-hand side of Eq. (12.57) is negative according to Eq. (12.55), this implies
that dD
(n)/dn < 0.
The intuition behind this result follows from the restriction on the joint distribution between
rm value and credit risk where high-quality rms have safer debt but are less valuable. As a
result of this restriction, insiders face opposing incentives in the equity and debt markets. If they
misrepresent the equity claims as being of high value, then the debt claims will be overvalued.
As a result of this trade-off, insiders have the incentive not to sell overvalued claims. Therefore,
low-quality rms will issue more debt than high-quality rms.
Brennan and Kraus [1987] extend Heinkels model and derive conditions under which the
adverse selection problem can be costlessly overcome by an appropriate choice of nancing
strategy. In Heinkels analysis, he takes the security types as given and demonstrates a fully
revealing equilibrium for a particular type of information asymmetry. In contrast, Brennan and
Kraus derive the properties the securities must have to be informative. The Brennan and Kraus
results are best illustrated by two examples they present in the paper.
In the rst example, consider a rm that has the opportunity to invest 10 at time 0. The
distribution of returns on the investment depends on the current state of the world, denoted by
A and B. The current state of the world is private information to the rm. If the rm does not
make the investment, it earns 100 or 140 at time 1, each with probability of 0.5. If the rm makes
the investment, the time 1 payoffs are 100 and 200 if the current state is A, and 80 and 195 if
the current state is B. As in the no-investment case, the probabilities are 0.5 each. The rm is
currently nanced with 100 in debt that matures at time 1 and 40 shares of equity. Since valuation
is assumed to be by expected value, the full information value of the rm with no investment is
120, with 100 being the value of debt and 20 the value of equity. Similarly with the investment, the
full information value of the rm is 150 in state A, with debt being worth 100 and equity valued
at 40. The corresponding gures for state B are 137.5, 90, and 37.5, respectively.
Consider the following fully revealing equilibrium. If state A occurs, the rm retires its debt at
its full information value of 100 and issues 110 shares of equity at the full information value of 1
per share. The true full information value of the old equity is 40. If state B occurs, the rm issues
10.67 shares of new equity at the full information value of 0.9375. The true full information value
of the old equity is 37.50 (=(40/50.67)(47.5)). For this equilibrium to be feasible, there should
be no incentive for the rm to misrepresent the state at time 0 by adopting the strategy for the other
state. Therefore, if state A occurs, the rm does not have an incentive to issue equity only since the
value of old equity would be (40/50.67)(50) or 39.47 < 40, the value with repurchasing debt and
issuing equity. Similarly, if state B occurs, the rm does not have an incentive to retire debt and
issue 110 in new equity since the original equity would be worth (40/150)(137.5) or 36.67 < 37.5,
the value with issuing equity only. Thus, in this case the rm repurchases debt and issues equity in
the better state and issues equity only in the worse state and the equilibrium is fully revealing. This
would imply that the announcement of a pure equity issue should be associated with a decrease in
stock value while a combination announcement of a debt repurchase and an equity issue should
be associated with an increase in stock value.
In the second example, assume that the density function for a rms cash ows is uniformly
distributed over (a , a +). At time 0, a is known to all participants, but is private information
to the rm. The rm has outstanding bonds that mature at time 1 with face value D
0
. The rm
can nance its capital needs by issuing junior subordinated convertible debt with face value D and
conversion ratio . Brennan and Kraus showthat, under certain conditions, a revealing equilibrium
is possible where the rm type that investors can infer from the nancing (, D) is
(, D) =
_
_
a D
0
_
2
+D
2
(1) /
2
_
. (12.58)
Equation (12.58) implies that higher conversion ratios and lower convertible bond face values are
associated with safer rms since / < 0 and /D > 0.
Constantinides and Grundy [1989] consider a three-date model where rms have assets in place
at time 0 and the only claim to the rm is common stock with management owning a fraction of
the outstanding equity. Management also announces the planned investment and how it will be
nanced at time 0. The nancing instrument is issued at time 1, matures at time 2, and can be
equity, a straight bond, a convertible bond, or some combination. If the amount the rm raises
is more than the investment required, the rm repurchases stock from the outside shareholders.
The rms value becomes common knowledge at time 2. At time 0, the rm value is a random
variable with a density function that depends on the investment and a parameter, , that represents
the information that the management has at time 0 and the market does not. The market has a prior
distribution for and updates its beliefs using Bayess rule based on the investment and nancing
announced by management.
In the case that the investment amount is xed, Constantinides and Grundy show that there is
a separating equilibrium that is fully revealing in which all rms make the investment and issue
convertible debt. The convertible debt issue is sufcient to cover the investment and repurchase
of some of the existing equity. The repurchase assures that the management has no incentive
to overvalue while the convertible debt issue removes the incentive to undervalue. If different
rms are allowed to have different optimal investment amounts and if investment is observable,
Constantinides and Grundy show that rms can fully separate using the investment amount and
the size of a straight bond issue with some share repurchase.
Noe [1988] considers a sequential game model of debt-equity choice in a two-date framework.
In his model, at time 0 the rm has assets in place that are expected to generate a time 1 cash
ow of x
1
. The rm also has access to a project that requires an investment (I) that is common
knowledge at time 0. The project yields a cash ow of x
2
at time 1. The quality of the rm is
dened by x = (x
1
, x
2
). Insiders know the quality of the rm, and outsiders have a common
probability distribution f () over rm types. Noe considers two cases regarding the relation
between knowledge of quality and cash owsone in which knowledge of quality provides perfect
foresight on cash ows and another in which the cash ow is equal to the perfect foresight values
plus a zero-mean random variable. Insiders must raise funds through debt or equity to nance the
project. He presents an example with three rm types, low, medium, and high. The high type rm
is assumed to be much better than the mediumtype rm, and the probability of the rmbeing a low
type is the lowest of the three probabilities.
11
He shows that, in equilibrium, all three rms accept
the positive NPV projects, with the low and high type rms pooling and issuing debt to nance
the project while the middle type rm separates and issues equity. Investors correctly identify the
quality of the rm issuing equity, and consequently the equity issue by the medium type rm is
priced correctly by the market. In contrast, either security issued by the low-quality rm will be
overpriced since it will be pooled with the medium type rm (if it issues equity) and the high type
rm (if it issues debt). Since the high type rm is much better than the medium type rm, the
overpricing is more severe for debt, and the low type rm chooses to issue debt. For the high type
rm, either issue is going to be underpriced, with the underpricing of debt being less since the
probability of being a low type rm is small. Consequently, the high type rm chooses to issue
debt. Noes model suggests that some rms may actually prefer to issue equity over debt, and
issuance of equity is not necessarily associated with the lowest-quality rm.
Nachman and Noe [1994] derive necessary and sufcient conditions for the issuance of secu-
rities to be an equilibrium outcome of raising external capital. They show that debt nancing is a
pooling equilibrium outcome if and only if rm types are strictly ordered by conditional stochastic
dominance, a strong version of rst-order stochastic dominance.
In a recent paper, Heider [2002] presents a model in which combinations of debt and equity
can be used to convey credible information to the markets. Heider argues that the different results
across papers are special cases of his model by using specic parameter values to describe the
quality of the rm. Heiders model is best explained by an example he presents in his paper. In a
two-date framework, consider a situation in which a rm needs to raise 10 units of outside capital
at time 0 to invest in a project. The rmhas no nancial slack and cannot sell its assets in place that
are worth 100 units. There are two types of investment projects, 1 and 2. A type 2 project returns
nothing with a probability of 0.25 and returns 18 units with a probability of 0.75. A type 1 project
returns 0 with a probability of 0.15 and 13 units with a probability of 0.85. Thus the expected rate
of return for a type 1 project is 10.5% and that for a type 2 project is 35%.
Assume rst that there is no information asymmetry regarding project type. Thus if we nance
a type 1 project with debt, the fair-price repayment on debt would be 11.765 units since the
expected payment would be 10 [=(0.15)(0) +(0.85)(11.765)]. If the type 1 project is nanced
with equity, the rmwould have to issue 9.005%of the equity of the rmsince the expected payoff
11
For example, he assumes that the assets in place generate cash ows of 0.3, 1.5, and 20 for low, medium, and high type
rms, respectively. The corresponding probabilities of rm type are 0.001, 0.991, and 0.008, respectively.
on this would be 10 [=(0.09)(100 +(0.85)(13))]. The corresponding gures for a type 2 project
are 13.333 and 8.811%. The value of the rm is independent of nancing in this case. Specically,
regardless of the method of nancing, the value of the rm with the type 1 project is 101.05 and
that with the type 2 project is 103.5.
Now consider the situation in which the rm knows more about the investment projects than
outsiders do. Suppose that rms use debt to nance type 1 projects and equity to nance type 2
projects. In that case, when outsiders observe a debt issue, they infer that the rm is investing in
a type 1 project, and they are willing to accept a fair-price repayment on debt of 11.765. On
the other hand, when they observe an equity issue, they expect an equity stake of 8.811%. Firms
knowing this have an opportunity to sell overvalued claims. Specically, if a rm investing in a
type 2 project mimics the issue decision of a rm that invests in a type 1 project, they can issue
debt with a repayment of 11.765 rather than the 13.333 they would have to offer under conditions
of information symmetry. Similarly rms that invest in type 1 projects mimic the issue decision
of rms with type 2 projects by selling 8.811% of the equity rather than 9.005%. Since investors
know that rms have an incentive to sell overvalued claims, they would not accept the offered
claims, and this cannot be an equilibrium.
Now consider the reverse system of nancing where equity is used to nance type 1 projects
and debt is used for type 2 projects. Therefore for type 1 projects, the rm issues 9.005% of its
equity, while for type 2 projects it uses debt with a repayment of 13.333. In this situation a rmhas
no incentive to mimic, and they obtain the same prices as in the full information situation. Thus the
use of equity nancing for safer projects and debt nancing for riskier projects is an equilibrium.
Heider argues that the driving force behind the pecking order and underinvestment argument in
Myers and Majluf is that equity nancing of riskier projects by rms gives rms with less risky
projects the incentive to mimic the former.
In a more formal model, Heider [2002] considers rms with access to two types of projects,
T =1, 2, with both projects requiring an investment of I. The projects return X
T
with probability
p
T
and zero with probability
_
1p
T
_
with p
1
p
2
and X
1
X
2
. The net present value of all
projects is positive, that is, p
T
X
T
> I. Denote the relative difference in success probabilities as
=
_
p
1
p
2
_
/p
1
and the relative difference in returns as =
_
X
2
X
1
_
/X
2
. Therefore, if =0,
project 1 dominates project 2 by rst-order stochastic dominance (FOSD); project 2 dominates
project 1 by FOSD if =0. If , project 1 dominates project 2 by second-order stochastic
dominance (SOSD), while neither dominates by FOSD or SOSD if > . Finally, the projects
are mean-preserving spreads (MPS) if = . According to Heider, many of the papers discussed
above are special cases of his model since they consider particular combinations of and . For
example, he argues that Heinkels model ts into the situation where > , while Myers and
Majluf considered situations where =0. He also suggests that Brennan and Kraus in the rst
example above, Constantinides and Grundy, and Nachman and Noe consider situations in which
either =0 or =0, while the second example from Brennan and Kraus assumes MPS.
Heider shows that some nancing contracts give the outside investor the same payoff regardless
of the quality of the project. Specically he shows that these belief-independent contracts must
satisfy the condition
D =X
1
_
( )
(1)(1)
_
, (12.59)
where Dis the belief-independent amount of debt and is the belief-independent fraction of equity.
Equation (12.59) implies that when projects are MPS ( = ), the belief-independent contract is
pure equity. If =0, pure debt is the belief independent contract. On the other hand, if < , the
belief-independent contract involves debt repurchases. Heider also shows that if debt repurchases
are allowed, then there exists a continuum of separating equilibria in which (1) safer projects are
nanced with more equity and less debt than riskier projects, that is,
1

2
and D
1
D D
2
with at least one strict inequality each, and (2) a deviation with more equity than is interpreted as
coming fromrms with a type 1 project and vice versa. In addition, he shows that when repurchases
of debt are allowed there exists a unique pooling equilibrium in which all projects are nanced
with belief-independent combinations of debt and equity. Again, any deviations with more equity
than are interpreted as coming from rms with type 1 projects and vice versa.
Consider the equilibria derived by Heider relative to those derived in previous work. If =0
(the Myers and Majluf case), the belief-independent contract is pure debt, and there is a pooling
equilibrium on pure debt. As stated earlier, Heider suggests that Heinkels model is analogous to
his model with > and no repurchase of debt. Therefore his model would imply that riskier
rms would issue more debt than less risky rms. This is consistent with Heinkels argument
that more valuable (less risky) rms would issue less debt. In the rst example from Brennan
and Kraus discussed above, with FOSD ( = 0 or = 0), the separating equilibrium requires
rms to repurchase debt and issue equity, the result in the Heider model if =0. On the other
hand, if =0, the pooling equilibrium is pure debt, the result in Nachman and Noe. If =0,
the Constantinides and Grundy result corresponds to the separating equilibrium that requires a
repurchase of equity by rms with riskier projects. Specically, when =0, the belief-independent
equity contract is =0. Thus, if rms with less risky projects issue debt only, ones with more
risky projects have to repurchase equity. Finally, in the second example of Brennan and Kraus,
projects are MPS and convertible bonds nancing riskier projects require higher debt face value
and lower equity into which debt can be converted. Heider argues that this is analogous to his
separating result that with = (MPS), riskier rms are nanced with more debt and rms
with safer projects issue more equity and buy back debt. Therefore, Heiders analysis suggests
that even though various models in the literature on signaling with debt and equity seem to
have conicting implications, they can be reconciled by recognizing that they have a common
logic.
3. Summary of Signaling Theory
This section has provided an overviewof information asymmetry and signals that could potentially
be used by insiders to reduce this asymmetry. These signals can either be costly (i.e., have
exogenous costs associated with them) or be costless (i.e., have costs that are endogenous).
Examples of costly signals considered include the amount of the rms equity that is retained by an
entrepreneur, the amount of debt issued by the rm, the size of the dividend declared, the type of
nancing used for an investment, and the decision to split the stock. Examples of costless signals
considered include stock splits, the amount of equity issued or repurchased, and the type of debt
issued or repurchased. Regardless of the choice of the signaling mechanism, the equilibria achieved
from signaling can be of the separating kind, where high-quality rms cannot be mimicked by
low-quality rms, or the pooling kind, where outsiders cannot differentiate between the two types
of rms.
B. Agency Theory 439
B. Agency Theory
The separation of ownership fromcontrol is one of the basic tenets of a free-market society because
it allows specialization.
12
For example, Ms. Smith may have enough wealth to own a farmbut lacks
the skill to run it. Mr. Jones, a farmer with years of experience, may lack the wealth to own a farm
and can earn his highest income by operating a farm. Ms. Smith, the principal, can hire Mr. Jones,
the agent, and they can both be better off. But when ownership and control are separated, agency
costs arise. Ms. Smith wants to maximize the value of the land, which depends on its ability
to produce crops over a long period of time minus the cost of production (e.g., fertilizer, seed,
machinery, and Mr. Joness share of the crop each year) while Mr. Jones wants to maximize his
share of the crop. Agency costs include the cost of monitoring, losses due to the choice of objective
function, and informational asymmetries.
Agency problems arise in rms because corporate decisions are made by managers (agents)
on behalf of the rms capital suppliers (principals). In most agency models, the sequence of
events starts with the principal choosing the agents compensation system, which depends on the
performance measures that the principal species as well as the nal outcome, c(s, p), where s is
the nal outcome and p are the performance measures. Based on this contract, the agent chooses
an action a, for example, decisions of nancing and investment. This action along with some
exogenous (random) factors determines the nal outcome. Next, the performance measures p and
nal outcome s are observed, the agent is paid according to his compensation contract c(s, p), and
the principal gets to keep the difference between the nal outcome and the agents compensation,
s c(s, p).
Consider a principal whose utility function is dened as U (s c). The principal is assumed
to exhibit greed; that is, he has positive marginal utility of wealth (U
> 0). The principal is risk

neutral or risk averse; that is, he has either constant or decreasing marginal utility (U
0). The
principals utility is affected by the compensation paid to the agent both directly and indirectly.
The indirect effect is from the impact of the compensation function on the action chosen by the
manager, which in turn affects the distribution of the outcomes. Let f (s, p|a) represent the joint
probability distribution of outcomes and performance measures conditional on the agents actions.
Let us assume that both the agent and the principal have homogeneous beliefs regarding f (s, p|a).
Finally, denote the agents utility function as V(c) G(a).
In the remainder of the chapter, we tackle various principal-agent issues one at a time.
13
What
is the optimal contract when monitoring is the issue, that is, when
.
the agents actions are observable (no monitoring problem),
.
are not observable,
.
only the nal outcome can be observed,
.
the content can depend on both the nal outcome and other performance measures,
.
there are multiple actions by the agent?
12
The analysis and discussion in this section borrows heavily from Lambert [2001].
13
See Pendergast [1999] and Laffont and Martimort [2002] for excellent reviews of the literature.
Next, suppose information asymmetry between the principal and the agent is at the heart of the
problem. What is the optimal contract? Finally, we address corporate nance specic issues. How
does agency theory affect the mix of debt and equity, dividend policy, and investment decisions?
1. The Optimal Compensation Contract When the Agents Actions Are Observable
As a rst step in the analysis of this model consider a situation in which a incentive problem does
not exist.
14
In the aforementioned farm example, Mr. Jones, the farmer, has incentive to overfarm
the land, and therefore to destroy its value, because he receives a share of the crop, not a share
of the land value. In this framework, we would choose a compensation contract and an action
to maximize the principals utility subject to the agent achieving a minimum acceptable level of
utility, V. Specically, we can write the problem as
maximize
c(s,p),a
__
U
_
s c(s, p)
_
f (s, p|a) ds dp (12.60)
subject to
__
V
_
c(s, p)
_
f (s, p|a) ds dp G(a) V. (12.61)
The rst part of the above equation represents the expected utility of the principal for a given
compensation contract and action. The expectation is based on the utility of the payoff to the
principal, U
_
s c(s, p)
_
, and the joint density function for the nal outcome and the performance
measure subject to the action taken, f (s, p|a). The second part of the equation species that the
difference between the agents expected utility fromcompensation and that fromthe action chosen
has to be larger than the minimum level acceptable to the agent. Again, the expectation is over the
joint density function for the nal outcome and the performance measure subject to the action
taken.
If denotes the Lagrangian multiplier associated with the constraint, the problem can be
written as
maximize
c(s,p),a
__
U
_
s c(s, p)
_
f (s, p|a) ds dp +
___
V
_
c(s, p)
_
f (s, p|a) ds dp G(a) V
_
.
(12.62)
The rst-order condition for this optimization is
U
_
s c(s, p)
_
+V
_
c(s, p)
_
=0 or
U
_
s c(s, p)
_
V
_
c(s, p)
_ =. (12.63)
The above condition states that optimal risk sharing condition is such that the agents compensation
is set so that the ratio of the marginal utility of the principal to the marginal utility of the agent is
equal to a constant for all possible realizations of
_
s, p
_
. This solution is generally referred to as
the rst-best solution.
14
The incentive problem arises because the principal cannot observe the agents actions.
Equation (12.63) has a number of implications. For example, if the agent is risk neutral and
the principal is risk averse, Eq. (12.63) becomes U
_
s c(s, p)
_
=. Since is a constant, this
implies that marginal utilityU
is a constant. Therefore, the optimal compensation contract is for the

principal to bear no risk and conversely for the agent to bear all the risk. The optimal compensation
contract in this case would be c(s, p) =s k, where k is a constant. On the other hand, if the
principal is risk neutral and the agent is risk averse, Eq. (12.63) becomes V
_
c(s, p)
_
=
1
, and the
optimal compensation contract would be c(s, p) =k, with the principal bearing all the risk and
the agent bearing none.
When both the agent and the principal are risk averse, the optimal contract will incorporate
some risk sharing with the exact form of the risk-sharing function depending on the two utility
functions. For example, Wilson [1968] shows that if both parties have negative exponential utility
function, the optimal compensation contract is linear in the nal outcome s with a slope coefcient
equal to the ratio of the risk tolerances of the agent and the principal.
15
Specically, consider an
agent and a principal with utility functions given by U
i
(W) =
i
e
W/
i
, i =agent, principal,
and
i
is is risk tolerance. Substituting these utility functions in Eq. (12.63) yields the optimal
compensation contract
c = +
_
agent
_
s,
where =
agent
+
principal
and =

. Figure 12.1(a) provides a plot of this optimal compen-

sation function. As can be seen from this plot, the slope of the compensation function increases
with the risk tolerance of the agent; that is, the more risk the agent is willing to tolerate, the more
risk he bears in the optimal compensation contract.
Compensation contracts that deviate from the linear structure specied above can lead to
conicts of interest between the agent and the principal. For example, consider the compensation
contract in Fig. 12.1(b), which can be dened as
c(s) =
_
0
if s s
0
_
0
s
1
1
s
0
s
1
s
0
_
+
_
0
s
1
s
0
_
s if s
0
< s < s
1
1
if s s
1
. (12.64)
As can be seen from the gure, the agent receives a xed payment if the nal outcome is below
s
0
or above s
1
and receives a payment that is linearly related to the nal outcome if it is between s
0
and s
1
. Further assume that the outcome is measured and compensation paid at the end of a period,
while the agent expends effort over the period at a rate that can vary over the period. Consider the
situation in which the agent has expended effort over three-fourths of a period and is relatively
certain that the nal outcome will be above s
1
. In this case, the agent has an incentive to stop
expending any more effort since any outcome above s
1
provides him with the same compensation,
1
. On the other hand, if the outcome is close to (and below) s
0
, the agent has the incentive to take
large risks since his compensation cannot fall below
0
. This suggests that compensation contracts
that deviate from the rst-best solution can induce conicts of interest between the principal and
agent.
15
Recall from Chapter 3 that the negative exponential is U(x) =e
X/
, where is a measure of the individuals risk
tolerance.
Figure 12.1 (a) The
rst-best compensa-
tion contract when the
principal and agent
have exponential util-
ity, U(W) =e
W/
.
(b) An alternate compen-
sation contract.
Value of the final outcome
(a)
Equal risk tolerance
Agent with higher risk tolerance
C
o
m
p
e
n
s
a
t
i
o
n
(b)
C
o
m
p
e
n
s
a
t
i
o
n
AN OPTIMAL COMPENSATION CONTRACT WHEN THE AGENTS ACTIONS ARE
NOT OBSERVABLE Let us now assume that the agent will select actions that are in his best
interests given the compensation contract that is offered to him by the principal. As a link to
the discussion in the previous section, rst consider the situations under which the principal can
design a compensation contract that shares the risk optimally between the principal and the agent
and provides an incentive to the agent to select a rst-best solution action. If the agent is risk
neutral, the optimal risk-sharing contract is for the principal to receive a xed payment k and for the
agent to bear all the risk and receive the residual value s k. In this case the rst-best solution
action can be viewed as the solution to
maximize
a
U(k) +
___
V
_
c(s, p)
_
f (s, p|a)dsdp G(a) V
_
. (12.65)
Specically, this implies the rst-best solution action is chosen to maximize the agents expected
utility. Therefore, this action is identical to what the agent would select given his own incentives.
Thus, when the rmis owned by the agent because of the structure of the compensation contract,
he chooses an action that represents the rst-best solution.
The rst-best solution also obtains if the agent is risk averse and the principal can invert the
outcome function to infer the agents action. In this case, the principal can offer the agent a
compensation contract that contains the optimal risk-sharing terms if the rst-best action has been
selected by the agent and substantially penalize any deviation from the rst-best action.
Figure 12.2 A com-
pensation contract with
penalties for shirking
when the principal and
agent have exponen-
tial utility, U(W) =
e
W/
.
a* + a* a*
C
o
m
p
e
n
s
a
t
i
o
n
Penalty
Finally, rst-best solutions can also be achieved if the set of possible outcomes changes with the
action selected. For example, consider a situation in which the outcome is uniformly distributed
between [a , a +] if the agent picks action a. If the rst-best action is a
, then any realized

outcome that falls in the range
_
a
, a
+
_
is consistent with the agent having selected action
a
. This implies that if the nal outcome is below (a
), the principal knows with certainty

that the agent did not choose the rst-best action a
. Assume that the principal offers the agent a

compensation contract that contains the optimal risk-sharing terms if the nal outcome is in the
range
_
a
, a
+
_
and imposes a large penalty on the agent if the outcome is below (a
).
In this case, if the penalty is large enough, the agent will pick the rst-best action a
since he knows
that while he could put out an effort less than a
and have a good chance of the outcome being in

the range
_
a
, a
+
_
, there is some probability that the nal outcome will be below a
and he would suffer the penalty. Figure 12.2 provides an example of a compensation contract with
such penalties built in.
THE IMPACT OF CONTRACTING ON OBSERVABLE FINAL OUTCOME ONLY
Holmstrom [1979] considers a principal-agent model where the compensation contract depends
on the nal outcome only. The structure of the game is Stackelberg in that the principal moves rst
and makes a take it or leave it offer to the agent. If the agent accepts the contract, it is assumed
that he is able to select a level of effort (action) a belonging to a set of feasible actions A. The
outcome s is a continuous randomvariable that increases in the level of effort. The density function
for the outcome conditional on the effort is denoted by f (s|a). The density function is assumed to
have the property that if f (s|a) >0 for some effort a, then it is positive for all a A.
16
Finally, it is
assumed that the agents utility decreases with increasing effort, that is, G
(a) <0 and G
(a) >0.
In this scenario, the principals problem is
maximize
c(s),a
_
U [s c(s)] f (s|a) ds (12.66)
subject to
_
V [c(s)] f (s|a) ds G(a) V,
16
The condition ensures that the principal cannot infer the level of effort from the outcome.
while for the agent,
maximize
c(s),a
_
V [c(s)] f (s|a) ds G(a).
The difference between the above formulation and the formulation for the rst-best solution
is that in this problem a constraint is added where it is assumed that the agent selects an action
that maximizes his expected utility given the compensation contract offered by the principal (the
incentive compatibility constraint). To make this problem more tractable, the incentive compati-
bility constraint as stated above is replaced by the rst-order condition for the maximization by
the agent. Specically, the rst-order condition is
_
V [c(s)] f
a
(s|a) ds G
(a) =0. (12.67)

If is the Lagrangian multiplier for the acceptable utility constraint and is the multiplier
associated with the rst-order condition in Eq. (12.67), the principals problem can be written as
maximize
c(s),a
_
U [s c(s)] f (s|a) ds +
__
V [c(s)] f (s|a) ds G(a) V
_
+
__
V [c(s)] f
a
(s|a) ds G
(a)
_
. (12.68)
The rst-order condition for the above maximization is
U
[s c(s)] f (s|a) +V
[c(s)] f (s|a) +V
[c(s)] f
a
(s|a). (12.69)
The condition in Eq. (12.69) can be rearranged as
U
[s c(s)]
V
[c(s)]
= +
f
a
(s|a)
f (s|a)
. (12.70)
A comparison of Eq. (12.63), the condition for the rst-best solution, and Eq. (12.70) indicates
that if is nonzero, the rst-best solution is not achievable. Holmstrom shows that has to be
positive as long as the principal wants the agent to expend more than the minimal level of effort
possible. In this case, if the principal gives the agent an optimal risk-sharing contract, this does not
provide incentive to the agent to expend a high enough level of effort. By imposing more risk on the
agent, the principal provides more incentive to increase effort, but this increased effort decreases
the agents utility. To compensate for this decrease and maintain utility at the acceptable level V,
the principal has to offer a higher expected compensation to the agent. Figure 12.3 provides a
comparison of the rst-best contract and a potential solution to Eq. (12.70). As can be seen from
the gure, the contract has a higher intercept and a larger slope than the rst-best contract. The
higher slope imposes more risk on the agent, while the higher intercept provides a larger expected
compensation.
Equation (12.70) indicates that the shape of the optimal contract depends on the shapes
of the principals and agents utility functions and the shape of the likelihood ratio function
Figure 12.3 A compen-
sation contract when
only the nal outcome
is observable and the
principal and agent
have exponential utility,
U(W) =e
W/
.
First-best solution
Contract with outcome observable
C
o
m
p
e
n
s
a
t
i
o
n
Figure 12.4 A compen-
sation contract when only
the nal outcome is ob-
servable, the principal
is risk neutral, and the
agent has a utility func-
tion, U(W) =[1/(1
)](
0
+
1
W)
(1 )
.
Convex ( = 1/2)
Linear ( = 1)
Concave ( = 2)
C
o
m
p
e
n
s
a
t
i
o
n
f
a
(s|a)/f (s|a).
17
In general, a sufcient condition for the contract to be nondecreasing in the
outcome s is for f
a
(s|a)/f (s|a) to increase in s; that is, the outcome density satises the mono-
tone likelihood ratio property (MLRP).
18
On the other hand, a linear likelihood ratio function
does not guarantee an optimal contract that is linear in the outcome. For example, consider a risk-
neutral principal and the agents utility function is V(c) =[1/(1 )](
0
+
1
c)
(1 )
. In this case,
V
(c) =
1
/(
0
+
1
c)
, and Eq. (12.70) can be written as

_
1
1
_
_
0
+
1
c
_
= +
f
a
(s|a)
f (s|a)
. (12.71)
Solving Eq. (12.71) for the compensation function yields
c(s) =
1
+
_
1
_
_
1
1
_ _
+
f
a
(s|a)
f (s|a)
_
_
1
_
. (12.72)
Even if is positive and the likelihood ratio function is linear in s, the compensation function
can be concave, linear, or convex in s. Specically if 0 < < 1, the compensation function is a
convex function of s. It is linear if is 1 and concave if is larger than 1. Figure 12.4 provides
an example of concave, linear, and convex contracts.
17
f
a
(s|a)/f (s|a) is a likelihood ratio since it is equal to ln f (s|a)/a and the latter represents the term that is set equal
to zero to obtain the maximum likelihood estimate of a.
18
See Milgrom [1981] for more details.
THE IMPACT OF CONTRACTINGONTHE FINAL OUTCOME ANDOTHER PERFOR-
MANCE MEASURES Since contracting on outcomes alone can result in deviations from the
rst-best contract, this implies that the addition of other performance measures into the com-
pensation contract can improve utilities if they improve incentives and risk sharing. Consider a
situation in which the nal outcome s and an additional performance measure p are observable.
Let f (s, p|a) denote the joint density function for the nal outcome and the performance measure
conditional on the effort a. In this case, the principals problem can be written as
maximize
c(s,p),a
__
U
_
s c(s, p)
_
f (s, p|a) ds dp
+
___
V
_
c(s, p)
_
f (s, p|a) ds dp G(a) V
_
+
___
V
_
c(s, p)
_
f
a
(s, p|a) ds dp G
(a)
_
. (12.73)
The rst-order condition for the above maximization is
U
_
s c(s, p)
_
V
_
c(s, p)
_ = +
f
a
(s, p|a)
f (s, p|a)
. (12.74)
As with the case where only the nal outcome is observable, is greater than zero if both the nal
outcome and another performance measure is observable. This implies that the optimal contract
depends on the performance measure p if the likelihood ratio function f
a
(s, p|a)/f (s, p|a)
depends on p. Holmstromshows that the likelihood ratio function depends on p as long as the nal
outcome s is not a sufcient statistic for s and p with respect to the agents effort a. This follows
from the fact that if s is a sufcient statistic, then even though p contains information about the
effort a, it does not provide any information in addition to what is provided by the nal outcome s.
To generalize, Holmstroms informativeness condition suggests that any additional performance
measures will be a valuable addition to a compensation contract as long as other available variables
are not sufcient statistics for this measure regardless of the noise associated with the measure.
Banker and Datar [1989] take the Holmstrom informative condition a step further by deriving
conditions under which multiple performance measures can be aggregated into one measure,
with the compensation contract being based on this one aggregate performance metric. They
show that if the principal is risk neutral, then multiple performance measures can be aggregated
linearly if the joint density function of the performance measures conditional on the agents effort
a belongs to the exponential family of distributions.
19
In addition, they show that the weight
attached to each performance measure in the aggregation is proportional to the sensitivity of
the performance measure to the agents effort and the precision of the measure. This suggests
that performance measures that are more sensitive to a will be weighted more in the aggregate
measure of performance. In addition, more noisy measures would have lower weights as compared
to measures that are more precise.
19
As pointed out by Banker and Datar, the exponential family of distributions includes many common distributions such
as the (truncated) normal, exponential, gamma, chi-squared, and inverse gaussian.
THE IMPACT OF ALLOWING MULTIPLE ACTIONS BY THE AGENT As suggested by
Lambert [2001], Eq. (12.74) can easily be extended to a multiple-action framework. Specically,
the optimal contract will be the solution to the rst-order condition
U
_
s c(p)
_
V
_
c(p)
_ = +
1
f
a
1
(p|a)
f (p|a)
+
2
f
a
2
(p|a)
f (p|a)
+
. . .
+
n
f
a
n
(p|a)
f (p|a)
, (12.75)
where
i
is the Lagrangian multiplier associated with action a
i
, i =1, . . . , m; a represents the
set of actions; and p represents the set of observable performance measures (that could include
the nal outcome s). The results of Banker and Datar as related to linear aggregation continue to
apply to this case, with the weights associated with each performance measure being proportional
to the sensitivity of the performance measure to the agents effort and the precision of the measure.
The sensitivity of the performance measure is dened as the weighted average of the sensitivities
of the performance measure to each individual action [E(p
j
|a)/a
i
], where the weight is the
Lagrangian multiplier
i
. One major problem with this result in this multiple-action framework
is that, in contrast to the single-action framework, other features of the model other than the
sensitivities of the performance measure to each individual action and the precisions affect the
relative weights. This requires a solution to the Lagrangian multipliers and makes the model
intractable since these multipliers are generally difcult to solve for.
Holmstrom and Milgrom [1987] have proposed an alternate formulation to the multiple-action
problem that is more tractable. In this model, the agent is responsible for an m-dimensional effort
a =(a
1
, . . . , a
m
) that is not observed by the principal both at the individual and the aggregate
levels. The principal bases the agents compensation contract on K observed performance mea-
sures, p =(p
1
, . . . , p
K
). One of the performance measures may be the nal outcome, but it is
also possible for the nal outcome to be unobservable to the principal. The performance measures
are assumed to be normally distributed with the agents actions affecting only the means of the
distributions. The expected values of the nal outcome and the performance measures are assumed
to be linear functions of the agents efforts. Specically, the outcome function is
s =
m
j=1
b
j
a
j
+
s
, (12.76)
and the performance measures are
p
i
=
m
j=1
q
ij
a
j
+
i
for i =1, . . . , K, (12.77)
where
s
and
i
, i =1, . . . , K, are jointly normal with means of zero. Denote
ii
as the variance
of p
i
and
ij
as the covariance between p
i
and p
j
. In the above denitions of the outcome and
the performance measures, the b
j
s and the q
ij
s measure the sensitivity of the nal outcome and
performance measure p
i
to action a
j
, respectively.
The principal is assumed to be risk neutral, and the agents utility function is of the form
V(W) =e
W
, where W =c(p) G(a), is the coefcient of absolute risk aversion, and
G(a) =0.5
m
j=1
a
2
j
is the monetary value associated with effort a. Finally, the compensation contract is assumed to
be a linear function of the performance measures and is given by
c(p
1
, . . . , p
K
) =
0
+
K
i=1
i
p
i
. (12.78)
Based on these assumptions, it can be shown that the principals optimization problem can be
written as
minimize
0
,... ,
K
m
j=1
_
b
j

K
i=1
i
q
ij
_
2
+
_
_
K
i=1
K
j=1
ij
_
_
. (12.79)
In Eq. (12.79), the rst term reects the principals desire to pick weights in the compensation
contract that minimize the difference between the sensitivity of the nal outcome to agents actions
(b
j
) and the sensitivity of the compensation to the agents actions
K
i=1
i
q
ij
,
the congruity effect. On the other hand, the second term reects the desire to minimize the risk in
the agents compensation since the agent would have to be compensated more for bearing added
risk, the sensitivity-precision effect.
In the case where there are two performance measures, p
1
and p
2
, Datar, Kulp, and Lambert
[2001] show that the ratio of the weights of p
1
and p
2
in the compensation contract are
2
=
m
j=1
b
j
q
ij
m
j=1
q
2
2j

m
j=1
b
j
q
2j
m
j=1
q
ij
q
2j
+
12
m
j=1
b
j
q
2j

22
m
j=1
b
j
q
1j
m
j=1
b
j
q
2j
m
j=1
q
2
1j

m
j=1
b
j
q
1j
m
j=1
q
ij
q
2j
+
12
m
j=1
b
j
q
1j

11
m
j=1
b
j
q
2j
. (12.80)
The rst two terms in both the numerator and the denominator reect the congruity effect while the
second two terms reect the sensitivity-precision effect. Thus, if the agent is risk neutral ( =0)
or if the performance measures are noiseless (
ij
=0 for all i, j =1, . . . , K), then the principals
problem is to design a contract that makes the agents overall performance measure
K
i=1
i
p
i
as congruent as possible to the nal outcome s. On the other hand, if the agents overall performance
measure is perfectly aligned with the nal outcome, the model is essentially the same as the single-
action model where the weight assigned to each performance measure is related to the sensitivity
and noise of the measure.
2. The Impact of Information Asymmetry between the Agent and the Principal
Consider a situation in which an agent receives an information signal m with a probability density
function of g(m) and uses the signal to update the density function of the nal outcome and other
performance measures to h(s, p|a, m). Further assume that the agent receives the signal after
C. Agency Theory and Finance 449
signing his compensation contract but before selecting his action, that he can leave the rm after
observing the signal, and that he can communicate the signal to the principal. In this case the
principals problem can be written as
maximize
c(s,p,m),a(m),m(m)
E
s,p,m
_
U
_
s c(s, p, m)
_
| a(m)
_
subject to (for all m)
E
s,p|m
__
V
_
c(s, p, m)
_
G [a(m)]
_
|a(m)
_
V, (12.81)
where
a(m) =the a that maximizes E
s,y|m
_
V
_
c(s, p, m)
_
|a
_
G(a) for each m,
m(m) =the m(m) that maximizes E
s,y|m
_
V
_
c(s, p, m)
_
|a
_
G(a) for each m.
We dene m(m) as the agents message to the principal after observing the signal m, which by the
revelation principle is assumed to be the true signal m(m).
20
The revelation principle establishes
that by recognizing the agents rational behavior, one can restrict oneself, without loss of generality,
to the class of truthful messages. Specically, it assumes that the principal is able to credibly
commit to not opportunistically use the information revealed by the message and can get the agent
to truthfully signal his private information by promising the agent whatever he would have received
by lying.
In the above formulation, the optimal compensation contract is based not only on the nal
outcome s and the performance measures p but also on the message sent by the agent. The major
differences between this formulation and that presented earlier under conditions of symmetric
information are in the agents minimum acceptable utility (AU) and incentive compatibility (IC)
constraints. Specically in the symmetric information case, the AUconstraint is based on expected
utility, but in this case it is replaced by a set of AU constraints, one for each signal m. This
change is necessary to ensure that the agent is prevented from leaving after observing the signal
for all realizations of m. Similarly, the single IC constraint is replaced by a set of IC constraints,
one for each signal m. In general, these models demonstrate that the principal will be worse off
in the presence of private information since the agent is likely to earn rents from his superior
information.
21
C. Agency Theory and Finance
Agency theory has been applied to a number of different areas in nance. Although the economic
theory of agency is mainly focused on the structure of managerial compensation contracts that
mitigate agency problems, the nancial theory of agency also analyzes the impact of the conict
between managers and a rms claimholders and the conict between claimholders on issues
related to optimal levels of investment and risk bearing by the rm and optimal capital structure.
In this section, we discuss the nancial theory of agency by rst looking at the impact of manager-
claimholder conicts on rm value, investment decisions, and compensation contracts. We will
20
See Myerson [1979] for more details on the revelation principle.
21
See Sappington [1983] for more details.
then offer an analysis of stockholder-bondholder conicts and its impact on various nancial
decisions including capital structure and structure of investments by rms.
1. Conflicts between Managers and Stockholders
As stated by Jensen and Smith [1985], there are three main sources of conict between managers
and claimholders. The rst is the choice of effort by managers and has been discussed extensively
in the previous section. The second stems fromthe fact that since investment in rm-specic human
capital represents a signicant portion of the managers wealth, she is concerned about the total
risk of the rm even though the shareholder can diversify away most of that risk. As a result, a
manager may make investment decisions that help diversify the rm but may not be in the best
interest of shareholders. Finally, the third source of conict arises from the differential horizons
of the managers and claimholders. A managers claim on the rm is limited to their tenure with
the rm, but a rms life is innite. As a result, managers would tend to place less weight on cash
ows occurring after their horizon in making decisions.
Jensen and Meckling [1976] analyze the impact of the rst source of conict by comparing
the behavior of a manager who owns 100% of a rms equity with that of a manager who sells a
portion of the equity to outsiders. Specically, consider a rm in which X is a vector of activities
from which the manager derives nonpecuniary benets.
22
Assume that the present value of cost of
generating X is C(X), and the total dollar present value to the rm of the productive benet of X
is P(X). Therefore, the net dollar gain to the rm of X is V(X) =P(X) C(X).
23
The optimal
level of factors and activities X
that is picked by a manager that is a 100% owner is dened by

V
_
X
_
X
=
P
_
X
_
X
C
_
X
_
X
=0. (12.82)
Thus for any X > X
, the net dollar gain to the rm must be lower than that at the optimal,
that is, V(X) < V
_
X
_
. Dene the dollar cost to the rm of providing the increment X X
of
activities as F =V
_
X
_
V(X). Figure 12.5 provides a plot of the trade-offs between the value
of the rm, V, and the value of the nonpecuniary benets, F. In particular, the line VF denes
the possible combinations of V and F that are available to the rm and its owners. Since it has
been assumed that an increase of one unit in F causes a decrease of 1 unit in V, the slope of VF
is 1. Consider a manager whose utility function, U(V, F), is dened in terms of the indifference
curves U
1
, U
2
, and U
3
. If the manager owned 100% of the rm, the manager would pick a level of
F such that her indifference curve is tangent to the line VF. This is represented by A in the gure
and results in the value of the rm being V
. If the manager sells 100% of the rm to outsiders

and if the rst-best solution can be achieved, then the outsiders should be willing to pay V
for
the rm.
Assume that the rst-best solution cannot be achieved and the manager retains a fraction of
this rm. Further assume that the outsiders pay the manager (1)V
for their share of the rm.

Given the change in ownership, the cost to the manager of consuming 1 more unit of nonpecuniary
benets is no longer 1 unit. Specically, although the decision to consume more than one unit in
nonpecuniary benets reduces rm value by one unit, the cost to the manager is only since the
remaining (1) is borne by the outsiders. Thus the new (V, F) constraint faced by the manager
22
This corresponds to what has been previously referred to as the managers action or effort.
23
This assumes that X has no impact on the equilibrium wage of the manager.
Figure 12.5 The value of the rm (V) and the value of nonpecuniary benets consumed (F) when the
manager owns a fraction of the rm and has indifference curves denoted by U.
B
C
F* F
Value of nonpecuniary benefits
V
a
l
u
e

o
f

f
i
r
m
F
V
V
V
2
= V*
V
1
V
F F
2
F
1
A
U
2
U
3
U
1
is dened by a line that passes through A but has a slope of , the line V
1
F
1
in Fig. 12.5. If the
owner-manager is free to choose the level of nonpecuniary benets subject only to the loss she
incurs as a partial owner, she will move to point B, where the value of the rmwill fall to V
and the
value of the nonpecuniary benets consumed increases to F
. Since the outside investors are aware

of this incentive to increase consumption of nonpecuniary benets, they will not pay (1 )V
for their share of the rm. Specically, assume that the outsiders pay S for their share of the rm.
The wealth of the manager is then given by W =S +V(F, ), where V(F, ) represents the
value of the rm given a nonpecuniary benet consumption of F and managerial ownership of
a fraction of the rm. Based on this level of wealth, the manager will select a level of F at a
point where the (V, F) constraint faced by the manager is tangent to an indifference curve. In
addition, the value of F will also lie along the line VF. This implies that the (V, F) constraint
faced by the manager given that she has sold a fraction (1) of the rm for S will be given
by the line V
2
F
2
in Fig. 12.5. In this scenario, the managers optimal choice is denoted by the
point C, where the value of the rm is V
and the value of the nonpecuniary benets is F
. The
result that point C represents the point of tangency follows from the fact that, if the optimal point
is to the left of C, this implies that the outsider is paying less than the value of the claim he
acquires. On the other hand, if the optimal is to the right of C, the outsider is paying more than
the value of the claim. Therefore, if the outsider pays an amount equal to the value of the claim
he acquires, the solution has to be represented by point C and S =(1)V(F, ) =(1)V
.
This also implies that the manager bears the full cost of reduction in rm value since her wealth
is given by W =S +V(F, ) =(1)V
+V
=V
. The manager incurs a welfare loss in

this case since the decreased utility associated with the reduction in rm value is more than the
added utility associated with the consumption of additional nonpecuniary benets. The reduction
in market value represents the agency cost caused by the sale of equity to outsiders.
Figure 12.6 The optimal scale of the rm when the manager owns a fraction of the rm.
D
E
F
G
C
B
A
Value of nonpecuniary benefits
Expansion path with fractional ownership by manager
Expansion path with 100% ownership by manager
P
e
c
u
n
i
a
r
y

w
e
a
l
t
h
Jensen and Meckling also show that the agency problem associated with effort can impact
the optimal scale of investment by the rm. Specically consider an entrepreneur with initial
(pecuniary) wealth of W and monopoly access to an project requiring an investment of I and
value V(I), with the project being subject to diminishing returns to scale in I. Figure 12.6
provides a graphical solution to the optimal scale of the rm. In this gure, the total wealth of the
entrepreneur is plotted along the vertical axis, where total wealth is dened as W +[V(I) I],
while the horizontal axis plots the value of nonpecuniary benets. The market value of the rm is
a function of the level of investment and the value of the nonpecuniary benets consumed by the
manager and is denoted by V(I, F). Let V(I) denote the value of the rm when the nonpecuniary
benets consumed are zero. Consider potential combinations of the managers wealth and value
of nonpecuniary consumption by the manager for various levels of investment, I
1
, I
2
, I
3
, . . . , I
N
,
where I
N
represents the value-maximizing investment level, that is,
V(I)
I
=1
at I =I
N
. For example, at an investment level of I
3
, the potential combinations lie on a line joining
the point (W +[V(I
3
) I
3
]) along the vertical axis and point F
3
along the horizontal axis. At each
investment level, the manager chooses a level of nonpecuniary benets based on the tangency
between his indifference curves and the line representing potential combinations of the managers
wealth and value of nonpecuniary consumption by the manager. For example, for investment level
I
3,
the manager chooses to consume F
3
in nonpecuniary benets and have a pecuniary wealth
of W +[V(I
3
, F
3
) I
3
]. Therefore, the path ABCD represents the equilibrium combinations of
wealth and nonpecuniary benets that the manager would choose if he could nance all levels of
investments till I
N
.
Nowconsider the situation in which the manager can nance investments up to a level of I
1
only
and has to obtain outside nancing for all levels of investments above I
1
. Note that this implies
that the fraction of the rm that the manager retains decreases with the scale of the investment,
from 100% at an investment level of I
1
down to (I
1
/I
N
) percent for an investment level of I
N
.
If the rst-best solution could be achieved, then the expansion path would still be ABCD even
when the manager obtains outside nancing. If the rst-best solution cannot be achieved, then the
manager will choose to consume more nonpecuniary benets when outside nancing funds part
of the rms investments. The deviation from the rst-best solution will be larger the larger the
fraction of outside nancing. This follows from the argument that the larger the fraction of outside
nance the lower the cost to the manager of $1 of nonpecuniary benets. As a result, the manager
would choose an expansion path denoted by AEFG, where at each point his indifference curve is
tangent to a line with a slope equal to , the fraction of the rm owned by the manager. Since
decreases with increasing investment, the tangency point would move further away from the rst-
best solution as the level of investment increases. In Fig. 12.6, the manager maximizes his utility
at point F, where he consumes F
3
and the value of the rm is V(I
3
, F
3
) I
3
. Thus, the manager
chooses a suboptimal level of investment because of the agency problem associated with effort.
To see this more formally, consider the conditions under which the optimal investment level
is chosen if the rst-best solution can be achieved. Specically, the manager chooses the level of
investment at which
V(I)
I
=1.
If the rst-best solution cannot be achieved, the manager chooses an investment level I at which
V(I, F)
I
+
F
I
=1,
where =I
1
/I. Since V(I, F) =V(I) F, this implies that the investment level chosen satises
V(I)
I
(1)
F
I
=1
or
V(I)
I
=1+(1)
F
I
> 1
since the level of nonpecuniary benets consumption increases with scale, that is,
F
I
> 0. Since
the project is subject to diminishing returns to scale in I and the rst-best solution is a level of
investment I
N
that solves
V(I)
I
=1,
the level of investment I
that solves
V(I)
I
> 1
has to be such that I
<I
N
. This implies that the manager chooses a suboptimal level of investment.
As stated earlier, since investment in rm-specic human capital represents a signicant portion
of the managers portfolio, she is concerned about the total risk of the rm even though the
shareholder can diversify away some of that risk. Reagan and Stulz [1986] provide a detailed
analysis of the risk-sharing incentives between shareholders and employees (managers). Consider
a manager who earns a compensation c and holds nontradeable assets with end-of-period value of
W.
24
The managers expected utility is given by
E(V) =aE(W +c) bVar(W +c) =a [E(w) +E(c)] b [Var(W) +2Cov(W, c) +Var(c)],
where E(W) is the expected value of the asset, E(c) is the expected compensation, Var(W) is
the variance of the asset, Var(c) is the variance of compensation, Cov(W, c) is the covariance of
the asset, and compensation and a and b are positive constants. The objective of shareholders is
to maximize the market value of their share of rm revenues (s). The compensation contracts of
the managers are assumed to be linear in revenues, that is, c = +s (, 0), and managers
require that the expected utility from compensation received is at least as high as what they could
obtain from other alternatives, that is, E(V) V.
25
Based on these assumptions, the capital asset
pricing model suggests that the present value of the shareholders position in the rm is given by
S =
(1)E(s) (1)Cov(s, R
M
)
1+r
f
, (12.83)
where E(s) is the expected revenue, R
M
is the return on the market portfolio, r
f
is the risk-free
rate of return, is the market price of risk
E(R
M
) r
f
r
m
,
and Cov(s, R
M
) is the covariance between revenue and the market return.
The maximization problem faced by the shareholder is
maximize
,
(1)E(s) (1)Cov(s, R
M
)
1+r
f
(12.84)
subject to the constraint
a
_
E(W) + +E(s)
_
b
_
Var(W) +2Cov(W, s) +
2
Var(s)
_
V. (12.85)
If denotes the Lagrangian multiplier associated with the constraint, the problem can be
written as
maximize
,
(1)E(s) (1)Cov(s, R
M
)
1+r
f
(12.86)
+
_
a[E(W) + +E(s)] b[Var(W) +2Cov(W, s) +
2
Var(s)] V
_
.
The rst-order conditions for this maximization are
a
1
1+r
f
=0 (12.87)
24
For example, this nontradeable asset could be the managers rm-specic human capital.
25
The disutility associated with effort is ignored in this analysis since the level of effort is assumed to be given.
_
aE(s) 2b (Cov(W, s) +Var(s))
_
E(s) Cov(s, R
M
)
1+r
f
=0 (12.88)
a
_
+E(s)
_
b
2
Var(s) V =0. (12.89)
Substituting for from Eq. (12.87) into Eq. (12.88) yields the following expression for :
=
aCov(s, R
M
)
2bVar(s)
Cov(W, s)
Var(s)
. (12.90)
Equation (12.90) implies that if the manager owns no risky asset other than their contract,
the optimal risk-bearing coefcient () depends on the market price of risk and the systematic
risk of the rm. Thus if the market price of risk is zero or if the rm has no systematic risk, the
managers bear no risk. Conversely, under these two conditions, the shareholders bear all the risk.
If the manager does own risky assets, then the optimal risk-bearing coefcient depends on the
covariance between the assets and the rm. If the covariance between the two is negative, the
asset holdings act as a partial hedge against compensation risk and the manager is willing to take
more risk in the compensation contract. On the other hand, the opposite is true if the asset held by
the manager is positively correlated with the rm, for example, the asset is rm-specic human
capital. Thus, in this situation the manager is less willing to bear compensation risk. It is, therefore,
consistent with this model that the managers would be concerned about the total risk of the rm
and they would attempt to diversify the risk associated with their contracts by diversifying the rm
itself.
Finally, another source of conict arises from the differential horizons of the managers and
claimholders. This conict can also result in managers investing at levels below the optimum. To
see this, consider the following modied example from Jensen and Meckling [1979]. Consider a
rm in which the manager invests (his human) capital in the rm and receives nontradeable claims
on the cash ows contingent on employment. Let us assume that the opportunity cost faced by the
manager is i per annum and the expected tenure of the manager is T . Consider a perpetual project
with a cash ow normalized to 1 and an annual rate of return of r. The value of the project to the
rm is
1
r
. In contrast, the value of the project from the managers point of view is
_
1
i
_
1(1+i)
T
_
_
.
This implies that a manager will be indifferent to investing in the project or not investing if the rate
of return from the project is such that its value is equal to the value from the managers viewpoint.
This implies that given a tenure of T and an opportunity cost of i, the manager will use a hurdle
rate of r
for project accept/reject decisions, where r
is given by
r
=
i
1(1+i)
T
. (12.91)
Equation (12.91) indicates that the hurdle rate used for the project decreases with tenure, with the
limit being the managers opportunity cost when the tenure becomes innite. Table 12.1 provides a
numerical example for the relation between the tenure of the manager and the hurdle rate employed
for an opportunity cost of 10% (= i). As can be seen from this table, the shorter the tenure,
the higher the hurdle rate. This implies that managers will tend to underinvest because of their
Table 12.1 The Relation between Manager Tenure and Project Hurdle Rates for an Opportunity Cost
of 10%
Manager Tenure (T years) Project Hurdle Rate (r* percent)
2 58
5 26
10 16
15 13
20 12
40 10
shorter horizon as compared to shareholders and the problem of underinvestment gets attenuated
the shorter the time horizon of the manager.
Rational shareholders will recognize the incentives facing managers to shirk, diversify, and
underinvest. Therefore, they would forecast the potential impact of these decisions and incorporate
it into the value they attach to the stock. Therefore, the rm would suffer losses from these
decisions, and these losses would represent the agency costs of outside equity nancing. These
agency costs of outside equity would have an impact on the amount of equity nancing a rm
chooses to use. Chapter 15 contains a detailed description on the relation between the agency
costs of equity and the capital structure of the rm.
2. Conflicts between a Firms Bondholders and Stockholders
Conicts arise between the bondholders and stockholders when managers make decisions that ben-
et stockholders at the cost of bondholders. Smith and Warner [1979] identify four major sources
of conict between these two claimholdersdividend payout, claim dilution, asset substitution,
and underinvestment. Dividend payout becomes a source of conict if a rmunexpectedly changes
its dividend payout and nances this increase by reducing the asset base or by reducing planned
investments. If a rmchooses to issue newdebt with equal or higher priority than existing debt, the
claimof the existing debtholders is diluted. Prior to the newissue of debt, the existing debt had sole
priority in its claim on the rms assets and revenues. With the new debt issue the existing debt has
to share the claim, and this causes a reduction in debt value. Asset substitution reduces the value
of debt by making it more risky. Specically, if a rm decided to substitute a high-risk investment
for a low-risk investment, the risk faced by debtholders increases, and this results in a reduction
in debt value. Finally, consider a situation in which a rm can potentially invest in a positive net
present value project with the benets of the investment accruing to bondholders. Under certain
circumstances, a rm may choose to pass up on this project (underinvest), thus causing bonds to
suffer an opportunity loss.
To illustrate the conict of interest arising from the dividend payout decision, consider a rm
that has assets in place that are valued today at $220. $20 of these assets are in cash (with a return
of 0%), while $200 is the value of a project that is expected to return 20% or 10% with equal
probability in one period. Assume that the rm has debt that matures in one period and has a face
value of $200 and pays no dividends to its stockholders. In this scenario, debt is riskless and will
be valued at $200. Specically, if the project returns 20%, the value of the rms assets are $260
($20 in cash and $240 from the project) and debt is paid its face value of $200. On the other hand,
if the project returns 10%, the rms assets are valued at $200 and debt is paid in full. Now
consider the scenario in which the rm decides to pay a dividend of $10 to shareholders from its
cash holdings. This implies that the assets in place are now valued at $210 with potential payoffs
of $250 and $190 next period. If the value of the rm is $250 next period, the bondholders still
get paid in full. On the other hand, if the rm is valued at $190, the payoff to the bondholders
is $10 less than the promised amount of $200. Since debt was riskless before the decision to
pay the dividend, this implies that the change in dividend payout causes a reduction in the value
of debt.
A more general way of viewing the impact of changes in dividend policy is based on the insight
in Black and Scholes [1973] that corporate securities can be viewed as options on the assets of the
rm. Denote V
t
as the value of the rm at time t . Assume that the rm consists of equity and one
issue of zero-coupon debt that has a face value of D and a maturity of T . Further assume that there
are no dividends paid to stockholders. This implies that the maturity date payoff on debt will be
given by
B
t
=
_
V
t
if V
t
< D
D if V
t
D
=D
_
D V
t
if V
t
< D
0 if V
t
D
(12.92)
Equation (12.92) indicates that the payoff on risky debt can be written as a combination of the
payoff on risk-free debt and a written put option on the rms assets with the exercise price and
maturity of the option being the face value and maturity of debt, respectively. Thus the value of
risky debt would be the value of risk-free debt minus a put option on the rms assets with an
exercise price of D and a maturity of T , that is,
B =
D
(1+r
f
)
t
P(V, D, T, r
f
,
V
),
where P() is the value of the put option, r
f
is the risk-free rate of interest, and
V
is the volatility
of returns on the rm. An unexpected increase in the dividends paid by the rm can be viewed as
a decrease in the current value of the rms assets. The impact of a decrease in rm value on debt
can be determined by examining the partial derivative of B with respect to V. Specically,
B
V
=
P
V
.
Since put values decrease with increases in the value of the underlying, this implies that
P
V
< 0
and
B
V
> 0.
Therefore, the value of a risky bond increases with rm value. This implies that the value of the
bond would decrease with decreases in rm value caused by unexpected increases in dividends
paid.
To illustrate the claim substitution problem, consider again the bond with a payoff described in
Eq. (12.92). As stated earlier, the value of the bond is
B =
D
(1+r
f
)
t
P(V, D, T, r
f
,
V
).
Now assume that a rm issues new debt with the same priority as existing debt, a face value of
dD, and a market value of dB. Since the debt is risky, we would expect the market value to be less
than face value, that is, dB < dD. Since the new bond has the same priority as the old debt, the
total market value of the debt in this rm is
B
=
D +dD
(1+r
f
)
t
P(V +dB, D +dD, T, r
f
,
V
),
and the new value of the existing debt is
B
NEW
=
_
D
D +dD
_
B
=
D
(1+r
f
)
t

_
D
D +dD
_
_
f
,
V
)
_
.
Since option prices are homogeneous in the value of the underlying and the exercise price, the
value of the put option,
_
f
,
V
)
_
can be written as
(D +dD) P
_
V +dB
D +dD
, 1, T, r
f
,
V
_
.
Substituting this expression for the value of the put in the new value of existing debt provides the
following expression for its value:
B
NEW
=
D
(1+r
f
)
t
(D)
_
P(
V +dB
D +dD
, 1, T, r
f
,
V
)
_
. (12.93)
Equation (12.93) indicates that the change in value of existing debt is given by
B
NEW
B =D
_
_
P(
V +dB
D +dD
, 1, T, r
f
,
V
)
_
+P(
V
D
, 1, T, r
f
,
V
)
_
=D
_
P
X
+P
Y
_
. (12.94)
Equation (12.94) suggests that the change in the value of existing debt is proportional to the
difference in the values of two put options (call them X and Y) that are identical in all respects
except for the values of the underlying. Put option X is valued for an underlying value of
V +dB
D +dD
,
while put option Y is based on an underlying value of
V
D
. Since dB < dD and put values decrease
with increases in rm value, put X is worth more than put Y and D
_
P
X
+P
Y
_
< 0. Therefore,
the value of existing debt falls with the issue of newdebt with equal priority, that is, B
NEW
B <0.
The intuition behind this result is that the new claim shares payoffs with the existing claims when
the value of the rm is below the promised payment. If the sharing rule is such that the existing
debtholders get less in every state of the world where the value of the rm is below the promised
payment, then the value of debt would fall.
The asset substitution effect follows directly from examining the impact of change in volatility
on the value of risky debt. In particular, a change in the volatility of returns on the rm would
induce the following change in the value of debt:
B
V
=
P(V, D, T, r
f
,
V
)
V
. (12.95)
Since option values increase with volatility, this implies that
P
V
is positive and
B
V
is negative.
Thus, a decision to substitute less risky with more risky assets results in a decrease in bond value
and, conversely, an increase in stock value.
The asset substitution effect can also be illustrated with a simple example. Consider a rm that
is currently planning to invest in a project with end-of-period payoffs of $200 and $180 with equal
probability. The rmis nancing this investment with a debt issue with a face value of $170. Given
the expected payoffs on the project and the promised payment to debt, holders of this debt would
expect $170 in each state of the world. Now assume that after issuing the debt, the rm switches to
another project with end-of-period payoffs of $230 and $150 with equal probability. The payment
to debt would no longer be riskless since the bonds would receive either $170 or $150 with equal
probability. Therefore, bondholders incur a loss of $30 with a 50% probability, and this would
result in a reduction in bond value. In contrast, if shareholders are the only other claimants in this
rm, they would now receive an extra $30 with a 50% probability, and this would result in an
increase in stock value that exactly offsets the decrease in bond value.
Myers [1977] provides a detailed analysis of the underinvestment problem. Consider an all-
equity rm with no assets in place and one future investment opportunity. The rm has to decide
whether to invest I one period from now (at t =1). If the rm invests, it obtains an asset worth
V(s) at t =1, where s is the state of nature that occurs at t =1. Thus at time t =1, the value of
the rm is V(s). Assume that for states s < s
a
, the value of the investment is less than the amount
of investment, that is, V(s) < I. Therefore, the rm will not invest in these states, and the value
of the rm at time t =0 is given by
V =
_

s
a
q(s)[V(s) I] ds, (12.96)
where q(s) is the value today of a dollar delivered at t =1 if and only if state s occurs. Since this
is an all-equity rm, the value of equity is equal to the value of the rm, that is, V
E
=V.
Now consider the situation where the rm can issue risky debt with a promised payment of D
and debt matures before the investment is made. The proceeds of the debt issue are used to reduce
the required initial equity investment. In this case, if the value of the investment is above the face
value of debt, equity holders would pay off the debt and make the investment. This follows fromthe
fact that if V(s) I > D, then by paying the debtholders D and making the investment of I, the
stockholders are left with V(s) I D >0. If s >s
b
denes the states for which V(s) I >D,
the value of equity would be given by
V
E
=
_

s
b
q(s)[V(s) I D] ds. (12.97)
On the other hand, if V(s) I D, the stockholders will turn the rmover to the bondholders.
The bondholders will choose to exercise the investment option as long as they receive positive value
from it. Therefore, the bondholders will invest as long as V(s) I > 0. The value of the bonds
will be given by
V
D
=
_
s
b
s
a
q(s)[V(s) I] ds +
_

s
b
q(s)D ds. (12.98)
The total value of the rm in this scenario can be obtained from the sum of Eqs. (12.97) and
(12.98) and is identical to that dened in Eq. (12.96). Therefore, in this case the existence of debt
nancing has no impact on the rm.
Now assume that the debt matures after the investment decision. Consider the situation where
the value of the investment is positive but is less than the promised payment to the debtholders,
that is, 0 < V(s) I < D. In this case, the shareholders would not make the investment since
their payoff from the investment is negative after accounting for the payment to debtholders, that
is, V(s) I D <0. Thus, the shareholders will forego the investment in all states belows
b
, and
the value of the rm be
V =
_

s
b
q(s)[V(s) I] ds. (12.99)
This implies that the existence of debt that expires after the time the rmhas to make an investment
decision can result in an incentive to underinvest.
The fact that stockholders face these incentives to expropriate debt value would be recognized
by rational bondholders. As a result, they will forecast the impact of these potential decisions and
incorporate them into their pricing decision. Therefore, the rm would suffer losses from these
decisions, and these losses would represent the agency costs of debt nancing. These agency costs
of debt would have an impact on the amount of debt nancing a rm chooses to use. Chapter 15
contains a detailed description on the relation between the agency costs of debt and the capital
structure of the rm.
Summary
In this chapter, we have examined various principal-agent issues related to the design of the optimal
contract when monitoring is the issue. Specically, we have examined the structure of the contract
when the agents actions are observable (no monitoring problem), when the agents actions are
not observable, when only the nal outcome can be observed, when the content can depend on
both the nal outcome and other performance measures, where there are multiple actions available
to the agent, and when information asymmetry between the principal and the agent is at the heart of
the problem. In the case where there is no monitoring problem and both the principal and the agent
have exponential utility functions, we showed that the optimal compensation contract is linear in
the nal outcome with the slope of the function depending on the risk tolerance of the agent relative
to the principal. If only the nal outcome is observable, we show that the optimal compensation
contract is going to involve more risk taking being imposed on the agent to provide an incentive
for the agent to increase effort and a higher expected compensation to offset the impact of the risk. In
the case where the contract is based on multiple performance measures, the optimal compensation
Problem Set 461
contract is shown to have higher weights attached to performance measures that are more sensitive
to the action taken by the agent and are less noisy. Similar results are shown to hold when the agent
can take multiple actions.
Finally, we addressed issues specic to corporate nance. In particular we discussed howagency
theory affects the mix of debt and equity, dividend policy, and investment decisions. We argue
that the presence of agency conicts between managers and stockholders, and stockholders and
bondholders, impose costs that increase with the amount of debt and equity in the rm. This
suggests that there would exist an optimal mix of debt and equity that would minimize the overall
agency costs faced by the company. A more detailed description of the impact of agency costs on
capital structure is contained in Chapter 15.
It should be recognized that even though we have focused on agency theory and its relation
to optimal contracting and corporate nance, this theory has been applied to a number of other
areas, for example, accounting, insurance, and property rights. As stated in Lambert [2001],
agency theory has been used in the accounting area to answer two questions. First, how do
accounting and compensation systems affect managerial incentives? And second, how do these
incentive problems affect the design of these systems? Mayers and Smith [1981, 1982] argue that
the differing costs of controlling incentive conicts between residual claimants and managers,
and between policyholders and residual claimants lead to different ownership structures in the
insurance industry. The property rights literature focuses on how costs and rewards are allocated
among various participants in an organization and how the specication of these rights is affected
by contracting.
PROBLEM SET
12.1 In the context of the Ross [1977] model, assume that managers are paid 20% of the time 0
and time 1 values of the rm. Further assume good rms are worth 250, bad rms are worth 150,
and the risk-free rate is 10%. What is the minimum cost of false signaling that has to be imposed
on management to ensure that all managers signal correctly?
12.2 In the context of the Bhattacharya [1979] model, assume that the personal tax rate is 25%,
the penalty associated with a shortfall is 50%, the project cash ows are uniformly distributed over
(0, 500), and the appropriate discount rate is 20%. What is the optimal dividend and value response
to this dividend? What is the impact of changing the personal tax rate to 30%? Changing the cost
of a shortfall to 70%? Changing the discount rate to 40%?
12.3 In the context of the Stein [1992] paper, show that (1) a bad rm will not mimic a good rm
by issuing straight debt and (2) a medium rm will not mimic either a bad or good rm.
12.4 Assume that both the principal and agent have negative exponential utility and that the actions
of the agent are observable. What is the optimal contract?
12.5 Consider a risk-neutral principal and an agent with utility function
V(c) =
_
1/ (1 )
_ _
0
+
1
c
_
(1 )
.
Assume that the contract being provided to the agent is based on the nal outcome only and that
the outcome density satises the monotone likelihood ratio property (MLRP). What is the optimal
contract? Under what circumstances is the optimal contract linear in the nal outcome? Convex?
Concave?
12.6 Assume that the opportunity cost of capital is 25%. Provide a plot of the relation between
the hurdle rate used by a manager for projects and the managers tenure.
REFERENCES
Akerlof, G., The Market for Lemons: Qualitative Uncertainty and the Market Mechanism, Quarterly Journal
of Economics, 1970, Vol. 85, 488500.
Austen-Smith, D., and J. Banks, Cheap Talk and Burned Money, Journal of Economic Theory, 2000, Vol. 91,
116.
Banker, R., and S. Datar, Sensitivity, Precision and Linear Aggregation of Signals for Performance Evaluation,
Journal of Accounting Research, 1989, Vol. 27, 2139.
Bhattacharya, S., Imperfect Information, Dividend Policy and the Bird in the Hand Fallacy, Bell Journal of
Economics, 1979, Vol. 10, 259270.
, Nondissipative Signaling Structures and Dividend Policy, Quarterly Journal of Economics, 1980, Vol.
95, 124.
Black, F., and M. Scholes, The Pricing of Options and Corporate Liabilities, Journal of Political Economy, 1973,
Vol. 81, 637659.
Brennan, M., and P. Hughes, Stock Prices and the Supply of Information, Journal of Finance, 1991, Vol. 46,
16651691.
Brennan, M., and A. Kraus, Efcient Financing under Asymmetric Information, Journal of Finance, 1987,
Vol. 42, 12251243.
Constantinides, G., and B. Grundy, Optimal Investment with Stock Repurchase and Financing as Signals, Review
of Financial Studies, 1989, Vol. 2, 445465.
Cooney, J., and A. Kalay, Positive Information from Equity Issue Announcements, Journal of Financial
Economics, 1993, Vol. 33, 149172.
Copeland, T., Liquidity Changes Following Stock Splits, The Journal of Finance, March 1979, Vol. 34, No. 1,
115141.
Copeland, T., and M. Brennan, Stock Splits, Stock Prices and Transactions Costs, Journal of Financial Economics,
1988, Vol. 22, 83102.
Crawford, V. P., and J. Sobel, Strategic Information Transmission, Econometrica, 1982, Vol. 50, 14311451.
Datar, S., S. Kulp, and R. A. Lambert, Balancing Accounting Measures, Journal of Accounting Research, 2001,
Vol. 39, 7592.
Fama, E., and K. French, Testing Trade-off and Pecking Order Predictions about Dividends and Debt, Review of
Financial Studies, 2002, Vol. 15, 133.
Grinblatt, M., and C. Y. Hwang, Signaling and the Pricing of New Issues, Journal of Finance, 1989, Vol. 44,
393420.
Guedes, J., and R. Thompson, Tests of a Signaling Hypothesis: The Choice between Fixed- and Adjustable-Rate
Debt, Review of Financial Studies, 1995, Vol. 8, 605636.
Heider, F., Signalling with Debt and Equity, working paper, New York University, 2002.
Heinkel, R., A Theory of Capital Structure Relevance under Imperfect Information, Journal of Finance, 1982,
Vol. 37, 11411150.
Holmstrom, B., Moral Hazard and Observability, Bell Journal of Economics, 1979, Vol. 10, 7491.
Holmstrom, B., and P. Milgrom, Aggregation and Linearity in the Provision of Intertemporal Incentives,
Econometrica, 1987, 303328.
Jensen, M., A Theory of the Firm, Harvard University Press, Cambridge, Mass., 2000.
References 463
Jensen, M., and W. Meckling, Theory of the Firm: Managerial Behavior, Agency Costs and Ownership Structure,
Journal of Financial Economics, 1976, Vol. 3, 305360.
, Rights and Production Functions: An Application to Labor Managed Firms and Codetermination,
Journal of Business, 1979, Vol. 52, 469506.
Jensen, M., and C. Smith, Stockholder, Manager and Creditor Interests: Applications of Agency Theory, Recent
Advances in Corporate Finance, 1985, 93131.
John, K., and J. Williams, Dividends, Dilution and Taxes: A Signalling Equilibrium, Journal of Finance, 1985,
Vol. 40, 10531070.
Krasker, W., Stock Price Movements in Response to Stock Issues under Asymmetric Information, Journal of
Finance, 1986, Vol. 41, 93106.
Laffont, J., and D. Martimort, The Theory of Incentives, Princeton University Press, Princeton, N.J., 2002.
Lambert, R. A., Contracting Theory and Accounting, Journal of Accounting and Economics, 2001, Vol. 32, 187.
Leland, H., and D. H. Pyle, Informational Asymmetries, Financial Structure and Financial Intermediation,
Journal of Finance, 1977, Vol. 32, 371387.
Lemmon, M. L., and J. F. Zender, Looking under the Lamppost: An Empirical Examination of the Determinants
of Capital Structure, working paper, University of Utah, 2002.
Mayers, D., and C. Smith, Contractual Provisions, Organizational Structure, and Conict Control in Insurance
Markets, Journal of Business 1981, Vol. 54, 407434.
, On the Corporate Demand for Insurance, Journal of Business 1982, Vol. 55, 281296.
McNichols, M., and A. Dravid, Stock Dividends, Stock Splits, and Signaling, Journal of Finance, July 1990,
Vol. 45, 857879.
Milgrom, P., Good News and Bad News: Representation Theorem and Applications, Bell Journal of Economics,
1981.
Minton, B., and K. Wruck, Financial Conservatism: Evidence on Capital Structure from Low Leverage Firms,
working paper, Ohio State University, 2002.
Myers, S., Determinants of Corporate Borrowing, Journal of Financial Economics, 1977, Vol. 9, 147176.
Myers, S., and N. Majluf, Corporate Financing and Investment Decisions When Firms Have Information Investors
Do Not Have, Journal of Financial Economics, 1984, Vol. 13, 187221.
Myerson, R., Incentive Compatibility and the Bargaining Problem, Econometrica, 1979, 6174.
Nachman, D., and T. Noe, Optimal Design of Securities under Asymmetric Information, Review of Financial
Studies, 1954, Vol. 7, 144.
Noe, T., Capital Structure and Signaling Game Equilibria, Review of Financial Studies, 1988, Vol. 1, 331355.
Pendergast, C., The Provision of Incentives in Firms, Journal of Economic Literature, 1999, 763.
Reagan, P., and R. Stulz, Risk-Bearing, Labor Contracts and Capital Markets, Research in Finance, 1986, Vol. 6,
217231.
Riley, J., Competitive Signalling, Journal of Economic Theory, 1975, Vol. 10, 174186.
, Informational Equilibrium, Econometrica, 1979, Vol. 47, 331360.
Ross, S., The Determination of Financial Structure: The Incentive Signalling Approach, Bell Journal of
Economics, 1977, Vol. 8, 2340.
Rothschild, M., and J. Stiglitz, Equilibrium in Competitive Insurance Markets, Quarterly Journal of Economics,
1976, Vol. 90, 629650.
Sappington, D., Limited Liability Contracts between Principal and Agent, Journal of Economic Theory, 1983,
121.
Smith, C., and J. Warner, On Financial Contracting: An Analysis of Bond Covenants, Journal of Financial
Economics, 1979, Vol. 7, 117161.
Spence, A. M., Job Market Signaling, Quarterly Journal of Economics, 1973, Vol. 87, 355379.
Stein, J., Convertible Bonds as Backdoor Equity Financing, Journal of Financial Economics, 1992, Vol. 32,
321.
Wilson, C., A Model of Insurance Markets with Incomplete Information, Journal of Economic Theory, 1978,
Vol. 6, 167207.
Wilson, R., On the Theory of Syndicates, Econometrica, 1968, 119132.

CH 12

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

CH 12

Uploaded by

Copyright:

Available Formats

12

. The entrepreneur has information that allows her to

is the maximum amount of debt a type B

, investors perceive the rm to be of type B. For this to be

) as long as her compensation from signaling correctly is greater than her

), is the change in the liquidation value from

(t ). Given the assumptions that the project is perpetual, the dividend

(t ) and V(D) schedules.

A) and the net present value (NPV) of the

is the market price of old shares if stock is issued.

B|issue and invest) (12.33)

B|issue and invest) > 0, this implies that P

A|issue and invest) < a

A|issue and invest), a con-

in the condition dened

A|issue and invest) +E(

B|issue and invest).

A|do not issue

P(n, z) +P T (n, P). (12.37)

maximizes the wage as

P(n, z) =k[n +c(z)]

X(n) T (n) C, (12.47)

(n) is the optimal value of D for insiders of a rm of quality n as determined by

> 0). The principal is risk

is a constant. Therefore, the optimal compensation contract is for the

. Figure 12.1(a) provides a plot of this optimal compen-

, then any realized

. This implies that if the nal outcome is below (a

), the principal knows with certainty

. Assume that the principal offers the agent a

and have a good chance of the outcome being in

(a) <0 and G

(a) =0. (12.67)

, and Eq. (12.70) can be written as

that is picked by a manager that is a 100% owner is dened by

. If the manager sells 100% of the rm to outsiders

for their share of the rm.

. Since the outside investors are aware

and the value of the nonpecuniary benets is F

. The manager incurs a welfare loss in

for project accept/reject decisions, where r

You might also like